Home

CIFER - MATLAB Interfaces: Development and Application

1. string 60 characters string 8 characters nE string 8 characters FRE MIs J D IF y N ry N positive scalar 1000 cell array 5 of strings cell array 20 of strings cell array 5 of strings positive scalar array 5 ry N cell array of strings positive scalar array 7 yv N ty N Q 1a Wy pt 89 Appendix E Online Help for Analysis Utilities This Appendix contains the information that is displayed when the help functionname command is used in MATLAB Function cifrms Description This mex file allows the user to use the RMS utility of CIFER Function returns mean square value and root mean squared value as well as information about the frequency response if desired msv rms npts lowfreq highfreq rmsfrgq cifrms in varargin Inputs in input structure fields defined below Information on the details of the structure can be found with an empty call gt gt cifrms options name value pairs to set individual data fields optional e g 10 I Outputs rms optional output allows user to set up a template structure using this call with no inputs gt gt out cifrms kkk OR rms root mean squared value msv Mean square value or full range RMS if percentage of range is specified npts number of points in response lowfreq smallest value in response hi
2. Again it should be noted that care should be used when employing the O flag for inputs as response names are not assumed by the code Output information is stored in the same fashion as cifhq The first two variables contain the numeric values for the crossover characteristics and the last variable has the frequency range of the response gt gt gt gt gt gt gt gt XOdb Xn180db frng cifxover 0 name XVLATSWP FRE A0000 AIL P toscrn off disp X0db 0 0 disp Xn180db 27 8183 180 0000 54 7285 29 4783 180 0000 63 2869 30 1053 180 0000 52 2747 disp frng 0 1396 31 4159 cifarith This utility allows basic arithmetic to be performed on two frequency responses The results are saved to a new response file A very simple example of running the arithmetic function is shown below gt gt gt gt gt gt gt gt gt gt in cifarith in names in outname in outid XVLATSWP FRE A0000 AIL P XVLATSWP FRE A0000 AIL R ae cifarith in test arith resp from Matlab Response test written to the database xxx Arithmetic operation successful Minimum Frequency is 0 1396 Maximum Frequency is 31 4159 1000 Values in Output Response Units are in RAD 79 The template sets all values to the defaults found when the arithmetic utility in CIFER is first run For example all scale factors are set to 1 the operation is multiplication and th
3. and MATLAB The results compared identically between either set of cases created using the same program However comparison of any combination of cases from differing setup methods resulted in the minuscule numeric error When overlaid in a plot there was no visual difference in the plots and thus no significant discrepancy between the two methods of creating a case The origin of the error was not located but suspected to result from MATLAB s use of double precision The conclusion was that unless time permitted or more significant discrepancy between results was found the problem would be considered minor For the most part the errors occurred in 25 regions of low coherence where less weight would be assigned to the results regardless of small numeric error The short term solution was to document the existence of the issue as a warning to users 3 1 2 2 Retaining Structure of Code A major concern of the project was to create new code that would be relatively easy to maintain and modify In order for the functions created for the MATLAB CIFER interface to be maintainable by CIFER lead programmers the code was written to mimic existing Fortran code wherever possible In other cases such as the MATLAB M files code was made uniform both in structure and in naming conventions so as to facilitate both ease of use and ease of maintenance Several specific methods were employed in this endeavor First in dealing with the
4. eee eee cseeeseeereecneeceaecnaeenaeeees 36 Figure 4 5 Matlab to CIFER Comparison scsssssssssessesssssssssesssssssssscsssssesecsssssesessessseeseesees 36 Figure 4 6 CIFER to bode Comparison cassizssosscoiesxcinveossiensavsssssssoievsvesqvieendiadeneessdauvassaiebsvbtisbes 36 Figure 4 7 Feedback Block Diagram es essessseesseceseceseceseceeeeseesseeseneeeneeeaeecsaecaaecsaeenaeenaeenes 38 Figure 4 8 Error Channel Verification ce esessecssecsseceseceeceseeeseessceseneeeaeeeaeecsaecaaecsaeeeeeeseeeees 38 Figure 4 9 Broken Loop Roll Gain Phase Margin Results ccescessceseceseeeeeeeseecnseceaeceseenseeees 39 Figure 4 10 Roll Bandwidth for Lateral Stick Input to Roll Angle Response 40 Figure 4 11 Stick and Actuator Input Autospectra eee eeesecsseceseceseceseeeseceeeecaaecaaeceaecnaeenseenes 41 Figure 4 12 Cutoff Frequency Compared to Bandwidth Frequency ces eeceeeseesseceseceneeeeeeees 42 Figure 4 13 Shddow 200 TUAY ctico ctore ser eee nEs a Ea EE EERE EE aE itie 42 Figure 4 14 Roll Angle to Rate Comparison 0 eeesseeessecseceseceseceeeeeeeeeseceseesaaecaaecsaecnaeenseeees 45 Figure 4 15 Pitch Angle to Rate Comparison esesescecssecssecsseceseceeceseceseeeseecaaecaaecaecnaeeeaeenes 45 Figure 4 16 Longitudinal Velocity Perturbations 0 eile ceecesceeeeeeeeseeeeeseeeseeeaaecaecsaeenaeesaeenes 46 Figure 4 17 Lateral Velocity Perturbations 20 0 0 cecceseceseceseceseceseesseeseeeeese
5. call This information can also be found in Appendix D Users can initialize a mostly empty 69 structure by calling the functions with no inputs and a single output This template contains all the correct fields with arrays specified to the correct size and many fields set to a default value For instance the option in FRESPID to cross correlate controls is set to yes Fields that require case specific information such as the case name are not given default values These functions make extensive use of cell arrays to store string information and users unfamiliar with cell arrays should review the topic The following examples detail the basics of using the functions opening saving and running cases Details specific to individual functions will be covered in following sections In addition to the short examples presented inline with the text detailed examples of setting up and running a simple second order mass spring Damper system and the XVLATSWP case from the XV 15 sample database 703 are presented in Appendices H and I respectively This database should be provided with the installation of CIFER It is very important to keep in mind that the MATLAB calls can only save frequency responses into the CIFER database at this time Due to complications with suppressing prompts from CIFER for information the option to save frequency responses as files has not been implemented in the MATLAB command line Creating a te
6. AIL RUD m_in outputs 1 4 P R AY VDOT m_in winon 1 5 n m_in frcalc 1 4 EE Il oll save case to database misosa m_in 2 102 oe Set up blank composite case cin composite oe Fill in appropriate values c_in casename thename c_in comments Matlab created XVLATSWP case c_in casein thename c_in caseout thename c_in inpgm MIS c_in controls 1 AIL c_in outputs 1 4 c_in winon 1 5 c_in frcalc 1 1 P R AY VDOT 1 1 1 1 Save case into database composite c_in 2 oe Run both cases frespid XVLATSP2 3 misosa XVLATSP2 3 composite XVLATSP2 3 The following figure shows a plot of the original XVLATSWP case as created and run in CIFER overlaid by the same case set up and run from MATLAB There is no appreciable difference between the two results CIFER Control Case vs Matlab Run XVLATSWP a CIFER S Matlab 30F 4 iv 3 on oO E a 1 Q oO c 9 e ost J e 0 1 1 10 100 Frequency rad s 103
7. Function to open save and run a FRESPID case Function may be called with no arguments to return a template mostly blank structure out frespid id cmd options INPUTS id either frespid structure or string with case name Information on the details of the structure can be found with an empty call gt gt frespid cmd command to have function perform a process 1 open FRESPID case 2 save a FRESPID case 3 save case and run batch job options name value pairs to set individual data fields optional e g casename XVLATSWP OUTPUTS out return structure for frespid data structure A template structure can be returned using an empty call gt gt out frespid EXAMPLE CALLS out frespid returns a mostly blank template frespid structure out frespid TEST 1 opens case named TEST returns frespid structure out frespid TEST 2 casename TEST2 winlen 45 40 30 20 15 saves TEST as TEST2 and changes window lengths the structure is returned frespid new_struct 3 sends new_struct a new FRESPID case to batch nothing is returned NOTES A totally new case must be specified via a structure not by the desired new name i e frespid NEWNAME 2 will NOT save a blank case entitled NEWNAME into the database DESCRIPTION OF FIELDS IN INPUT STRUCTURE SCREEN 2 id casename FRESPID case name id comments FRESPID case identifier string 83 id controls id outputs id
8. bode Comparison 36 4 3 UH 60 Simulation The first in depth validation of the CIFER MATLAB interface was based on UH 60 simulation data provided in the Combined CIFER CONDUIT RIPTIDE Training course This course was designed to give engineers a brief introduction to the three programs developed and distributed by the Flight Controls Group of which CIFER is one The second program is CONDUIT which is designed to optimize a control system around a parametric model for a system The third program is RIPTIDE which is a simulation program that will allow users to fly systems modeled in CONDUIT Together the programs constitute a very powerful control systems design suite The course data was examined for crossover and bandwidth characteristics using CIFER utilities The reference values for these properties were already provided from CONDUIT analysis and were used as a check to ensure the correctness of the CIFER results The course manual provided closed loop data necessary for bandwidth analysis In order to investigate crossover characteristics it was necessary to generate additional simulated flight recordings in RIPTIDE of the feedback and error channels Only the roll channel was examined for this analysis As a first step in the analysis frequency response arithmetic CIFER utility 9 was used to confirm the consistency of data channels used for error and feedback Equations 4 2 and 4 3 show the rela
9. 040409 In this example the name of a frequency response is specified when the structure is created it assumes defaults for inputs such as source and input output integration selection These defaults are defined in Appendix F and can be accessed by an empty call to the function as mentioned above Single character inputs such as those for source are not case sensitive No outputs need be specified the function will print all the results to the screen A more complex call is shown below without the printed output and using name value pairs for input The name value pairs do not have to be specified in a particular order as long as they occur in pairs gt gt in cifrms gt gt name XVLATSWP FRE AOOOO AIL P gt gt scorrect 2 gt gt minfreq 10 gt gt maxfreq 30 gt gt msv rms pts min max cifrms in name name spower scorrect minfreq minfreq maxfreq maxfreq toscrn off This example specifies limits to the frequency range examined and sets the power of s correction factor to 2 The default values can be used for these variables by assigning the inputs a value of 0 The output variables store the information that results from the function call msv and rms store the mean square value and root mean squared value respectively The variable pts stores the number of points in the response and min and max store the minimum and maximum frequencies in the respon
10. 2 ANALYSIS ea te hies ds Seds NE od Ga beta ase tees ENE EE EE EEE SEESE A avis EEEE EEES 59 5 3 FUTURE WORK aretace noteiert ok EE EA E ta tenses Gotcha ce O KE amp soleus EEEE EEEE OE reu EEE A EVERE REEE Guts 60 BIBLIOGRAPHY es scccsssicvassssectersssceccscassctestessecsseseoassacodencasssneasesecdescesedb oastasesousesebeoansaccsessasesbetetacedencasesboans 61 APPENDIX A SCREEN LAYOUT COMPARTSON cccsssssscscscssscscsccssssssscssssseesscessessscsscesssssees 63 APPENDIX B HELP DOCUMENTATION FOR COMMAND LINE INTERFACE s0000 68 APPENDIX C ONLINE HELP FOR MAIN PROGRAMS eseeseesessessessessesessossesoossesoesoesessosoessesessose 83 APPENDIX D STRUCTURE FIELD SPECIFICS FOR MAIN PROGRAMS essessessosessoesessossesoee 88 APPENDIX E ONLINE HELP FOR ANALYSIS UTILITIES seesesoesseseosoesessossesoesoesessoesesssssesose 90 APPENDIX F STRUCTURE FIELD SPECIFICS FOR ANALYSIS UTILITIES ceseseee 96 APPENDIX G ONLINE HELP FOR SUPPORT FUNCTIONS ccsscsssssssssesscecsesssesseeseessees 98 APPENDIX H MASS SPRING DAMPER CASE EXAMPLE sessessessesessossessessesoessesessesoosoesesso 100 APPENDIX I XVLATSWP CASE EXAMPLE eseesessesseseosoesessossessessescesoesessossesoessesoesoesessesoessesesss 102 vii Table of Figures Fig r 1 1 Th Role of Systemi Iderien nian ae e a et iati 1 Figure 1 2 Frequency Sweep Example UAV Flight Data eeeeeeseeeeseeeseerresresresresresreeresresreee 3
11. 5 array 20 5 of strings array 20 of strings tive scalar array 20 5 Pod ome array 2 10 0 array 10 of strings N AI Ut tive scalar 1000 array 5 of strings array 5 of strings tive scalar array 5 scalar scalar Qa lt 4 cet C positive positive positive positive positive 19 1C ry 1y y scalar scalar scalar scalar Wy N N N 20 10 of strings i ww eo eo eo eoa eaa a 88 gt gt misosa Structure fields are casename caseid casein caseout source savdb plot outpts controls outputs winon frall frealc plotopt grid lrgplot plotdev pltminfrq pltmaxfrq gt gt composite string 8 characters string 60 characters string 8 characters string 8 characters Dp iB ryt N ry N positive scalar 1000 cell array 10 of strings cell array 20 of strings cell array 5 of strings ry N cell array 20 of strings positive scalar array 12 ryt N ryt N 19 on Ww T positive scalar array 5 0 5 positive scalar array Structure fields are casename caseid casein caseout inpgm source savdb plot outpts controls outputs winon winlen frall frealc plotopt grid lrgplot plotdev loyh string 8 characters
12. Fh Fh Fh ole H oP o oe Save the structure into the database frespid f in 2 Change the window sizes and turn them on oe frespid MASSSPRG 2 winlen 30 25 20 15 10 winon tat maxfft 125 125 125 125 125 Set up blank composite case in composite o Fill in appropriate values c_in casename thename c_in comments mass spring system c_in casein thename c_in caseout thename c_in inpgm FRE c_in controls 1 IN NOTE The file massspring CIFERTEXT was created for this example in order to run the example this file must be created 100 c_in outputs 1 c in frcalc 1 1 c in winon oe ee ee OUT 1 1 Save case into database composite c_in 2 Run both cases frespid MASSSPRG 3 composite MASSSPRG 3 The figure below shows the results from the analysis The response is very clean with a drop in coherence at the mode Both the MATLAB and CIFER results overlay closely There is a small difference due to machine precision however this error tends to be on the order of a thousandth of a percent or less CIFER anal aa Mag dB Phase deg o 08 V 0 6 0 4 0 2 0 1 1 10 100 Frequency rad s Coherence 101 Appendix I XVLATSWP Case Example This example shows the MATLAB commands used to fully set up and run the XVLATSWP sample case provided with installations of CIFER Assign a b
13. Figure 1 3 Doublet Example UAV Flight Data sesseseeeeseeessssreeresreeresrrsserrrsseestesreeresresresrresee 5 Figure 1 4 Example CIFER Sereettc c x5scczssscestecwsdie hte deapaecastentatvatdec ore Dealeeraheadctiucee 6 Figure 1 5 Mass Spring Damper Left XV 15 Right 0 eeecsneceeceeeeeeeeeeeeeaeeeaeeenaeenaees 9 Figure 1 6 NASA Sikorsky UH 60 RASCAL Left Shadow 200 TUAV Right c ceeseeeees 9 Figure 2 1 Example Coherence Plot ccsi cccescsciecseecceveeeseetees ann i E cothes ethegees E ea 11 Fis re 2 2 Example FRESPID Screens i scccsesctssstsscdestetsseathts ssebesessateeugtdades Eg REKE E eta e itih 13 Figure 2 3 Example RMS Prompts romeinse eeano eeren eaa e ie te niet ina 16 Figure 3 1 Name value Pairs and Structures eseeseeeeeeeseeeeseesesreerestesrestessesrtsseestesrestestessesersee 23 Figure 3 2 Percent Error Comparisons cesscescecssecsseceseceseceseeeseesseeseneeeaeeeaeessaecsaecaecnaeenaeenes 25 Figure 3 3 Comparison of Navigation Menus sccesceseceseceseeeseeeseeseeeeeseeeseecaaecaaecaecsaeesaeenes 30 Figure 4 1 XVLATSWP Validation Examples cecessecssecsseceseceseceseceseeeseecaaecnaecaecsaeenaeeees 34 Figure 4 2 SISO Mass Spring Damper System ee eeeesecssecsseceneceseceseeeeeeeeeeeeaecaaecaecnaeenseeees 35 Figure 4 3 Mass Spring Damper Simulink Block Diagram cece ceeceeeeeeeneeeeeeesecesecnaeenseeees 35 Figure 4 4 Mass Spring Damper System Input and Output
14. Fortran mex files the names of variables transferring data from MATLAB into the CIFER common blocks were made identical to the common block variables but with an x appended on the end This notation would allow easy identification of variables by programmers already familiar with the old Fortran code Second the variables were renamed when they were passed into the MATLAB workspace because the Fortran variables have names suited to programming that may not be meaningful to an engineering user Last the various Makefiles which create the Fortran mex files were condensed and grouped according to the utilities and programs they created Thus all the functions that facilitate FRESPID use are created using a single Makefile CIFER programs and utilities vary slightly in their method of interfacing with the user as described in Chapter 2 Some use a screen interface where the user fills in fields and advances using the function keys When the end of the screens is reached the screen programs may call additional programs that send information to be processed by CIFER Other programs use a 26 command line based interface that runs calculations and returns results as the user steps through via prompting The three main programs for this project FRESPID MISOSA and COMPOSITE when run in CIFER use the screen interface and have separate functions to store retrieve and compute information The MATLAB interface uses these same functi
15. The out_x variable retains the information about the new case gt gt out_f frespid XVLATSWP 2 casename TEST caseout TEST gt gt out_m misosa XVLATSWP 2 casename TEST casein TEST caseout TEST gt gt out_c composite XVLATSWP 2 casename TEST casein TEST caseout TEST Cases can also be specified using the appropriate structure as the first input Shown below are calls using frespid that mimic the above example In both examples the output name and input name for misosa and composite is also specified if it had not then the output name would still be XVLATSWP The MATLAB functions do not assume the user wants to change the output name If the output field is blank then the functions assume the output name to be the same as the case name gt gt out_f frespid XVLATSWP 1 gt gt out_f casename TEST gt gt out_f caseout TEST gt gt frespid out_f 2 Users should be careful when using old cases as a template If the number of controls outputs data files or other such parameters changes old values can remain in the arrays of both the MATLAB structure and in the CIFER database Therefore it is good practice to carefully zero out unused fields in arrays when making changes This will prevent unwanted information from being retained as changes are made Running a batch job 3 as second input Specifying a case to run can be as straig
16. are not inherently bound by the math that makes a simulation correct and it is worthwhile to confirm that kinematic laws still hold true for them Analyzing spurious results allows engineers to accurately correct flight data or fix the instruments in order to acquire hopefully more accurate data The flight data includes measurements for the rates and accelerations with the addition of alpha and beta measurements taken from a nose boom These allow consistency checks for Equations 4 7 and 4 8 Measurements of phi and theta were not included in the flight time history files and thus there is no benchmark with which to compare Equations 4 4 and 4 5 However as phi and theta can be reconstructed from p and q Equations 4 6 through 4 8 could be used to generate responses for the velocity perturbations Equation 4 6 was not used with the flight data as there would only be one source of u calculations for comparison However as alpha and beta were included in the time histories v and w could be calculated by two methods and those results compared to check the data consistency of the time histories Figure 4 19 shows v comparisons where the solid line is the calculation of v using yaw rate roll angle and lateral acceleration and the dotted line is v solved using the beta response for a forward velocity of 85 knots The coherence drops rapidly at a very low frequency just over 2 rad s however the trends of the two calculations still match up clo
17. for the type of conditioning desired and the second row stores the value The convention is that 1 denotes expansion of data 2 denotes decimation and 3 denotes filtering The second field condunit contains the units to use in the case of filtering The example below illustrates these variables gt gt in frespid gt gt in conditioning 1 2 1 2 3 2 4 25 gt gt in condunit 1 Hz The data in the example will be filtered at 4 Hertz and decimated to 25 Hertz These operations are lumped into one somewhat ungainly variable because of how CIFER internally processes the data The default value for condunit is Hertz so it need not be specified unless Radians or non dimensional values are used When returning conditioning output from CIFER to MATLAB the function also defaults the units to Hertz If cases created using conditioning are opened in CIFER there may appear to be a small error in the ten or hundred thousandth decimal place This is due to MATLAB using double precision while Fortran uses a combination of precisions Numerous checks have verified that the end results are the same 72 The variables that support screen 8 are analogous to the fields in that FRESPID screen The general convention is that if a window length is specified as zero then automatic calculations on it will not be performed However if any of the other related variables such as numbers of input and output po
18. k k k k k k k k k k k k k kk kk K Please enter a case name Aircraft Figure A6 MATLAB GUI Screen 1 Case Name Comments z Tapus progran nane Input Case name Input prefix g BROWSE XVLATSWP Output prefix Output to database Output to file Generate Plots Number of Output Points outputs anaes Enable Figure A7 MATLAB GUI Screen 2 65 Plotting Options Transfer fn magnitude Transfer fn phase Coherence Input auto spectrum Output auto spectrum Cross spectrum Error Heavy grid Large plot Plot output device Figure A9 MATLAB GUI Screen 4 66 Figure A10 MATLAB GUI Final Screen COMPOSITE case to load Select desired controls outputs Case Comments Window composite for Tiltrotor Figure A11 MATLAB GUI Two Data Loading Screens 67 Appendix B Help Documentation for Command Line Interface Introduction This document details the use of the command line functions included in the CIFER MATLAB interface developed by NASA Ames CIFER is a tool also developed by NASA Ames for system identification using frequency responses This document has been written assuming the user has background in using CIFER Any questions concerning the operation of CIFER should be directed to the appropriate CIFER user manual This interface is designed as a tool to allow users to run CIFER programs and utilities from the command
19. minimum frequency for mpc plot in mpcmax maximum frequency for mpc plot in mpcdev output device Q MS C omprs V er S creen T alaris P ostScript in lpcplt create a linear phase and coherence plot in lpcmin minimum frequency for lpc plot in lpcmax maximum frequency for lpc plot in lpcdev output device Q MS C omprs V er S creen T alaris P ostScript in lsfit perform least squares fit NOTE Must create linear phase and coherence plot in order to use least squares fitting in lslow lower fitting frequency in lsup upper fitting frequency in lscoh use coherence weighting in lsdev output device Q MS C omprs V er S creen T alaris P ostScript in toscrn turn printed screen output ON or OFF Function cifxover Description This function calls the second half of CIFER s utility 8 the crossover calculations XOdb Xn180db FRrng cifxover in options Inputs in input structure fields defined below Information on the details of the structure can be found with an empty call gt gt cifxover options name value pairs to set individual data fields optional e g save Y Outputs XO0db optional output allows user to set up a template structure using this call with no inputs and a single output gt gt out cifxover kkk OR kkk X0db Array with results for 0 deg crossovers Xn180db Array with results for 180 deg crossovers FRrng array containing
20. num pts frinfo cifhq in toscrn off gt gt disp words 180 deg frequency 29 478342 Rad sec DB Gain 63 286926 dB Linear gain 0 000685 Hz tt 135 deg BW freq 29 463261 Rad sec l DB Gain 61 210114 qdB Linear gain 0 000870 Hz tt 6 dB Bandwidth frequency 29 437544 Rad Sec Another 6 dB Bandwidth frequency 27 796064 Rad sec Another 6 dB Bandwidth frequency 27 782501 Rad sec Another 6 dB Bandwidth frequency 24 950647 Rad sec Another 6 dB Bandwidth frequency 24 890440 Rad sec Another 6 dB Bandwidth frequency 17 924093 Rad sec Another 6 dB Bandwidth frequency 17 860493 Rad sec TWICE 180 FREQUENCY NOT FOUND The above call has turned off the screen output using the toscrn field The variable num contains only the resultant frequencies and gains that are shown in the words variable Those values on the right side of the equals sign pts saves the number of points in the response and frinfo is an array containing the starting and ending frequency magnitude and phase of the response It also contains starting frequency and phase for reference if modified using correction factors for power of s gain phase shift or time delay The correction factors can be used with a slightly more complex call shown below gt gt name XVLATSWP FRE AOOOO AIL P gt gt corrections 2 5 90 0 5 gt gt cifhq 0 name
21. of taking measured data from a physical system and analyzing it to develop a mathematical model of that system This is an important aspect of control system design as it allows for the validation of simulated system models optimization of existing control systems and handling qualities specification compliance Figure 1 1 shows how system identification fits into a design cycle Assumptions gt modei simulation gt Proicte Aircraft Motion Understanding Identification Aircraft Motion Figure 1 1 The Role of System Id Starting at the top left of Figure 1 1 assumptions are made that result in some form of mathematical model which describes an aircraft The model can then be applied to a simulation that will predict the motion of the aircraft Once a physical model of the aircraft is constructed physical measurements can be made of its actual motion and responses to input System identification can then be used to extract a new mathematical model of the aircraft The new model can be compared to the old model and the assumptions used to create it for greater physical understanding of the aircraft s motion The mathematical model created through system identification can either be nonparametric or parametric Nonparametric models do not assume an order or structure They can exist either in the time domain as an impulse response or in the frequency domain as a frequency response Frequency responses are typically repr
22. output ON or OFF Function cifplot Description This function allows the user to call CIFER utility 19 from the Matlab command screen This utility generates a canned plot for frequency responses out ciplot in options Inputs in input structure fields defined below Information on the details of the structure can be found with an empty call gt gt cifplot options name value pairs to set individual data fields optional e g source F Outputs out optional output allows user to setup a template structure using this call with no inputs gt gt out cifplot DESCRIPTION OF FIELDS IN INPUT STRUCTURE SCREEN 1 in array 5 which arrays l mag 2 phas 3 coh 4 GXxX 5 GYY 6 GXY 7 err 8 pcoh2 9 pcoh3 10 pcoh4 11 pcoh5 12 pcoh6 13 pcoh7 14 pcohs8 15 pcoh9 16 mcoh in correct 5 use corrections 0 to skip 1 to apply in names 5 response names in gain 5 gain correction in phase 5 phase shift correction in spower 5 s power correction in source SCREEN 2 in in in in in in an in Land_port in in in pminfrg pmaxfrq pmin 1 5 pmax 1 5 pinc 1 5 device grid thick xaxis yaxis source of data F File D Database plot min freq 0 for default plot max freq 0 for default plot y axis mins plot y axis maxs plot increments output device Q MS C omprs V er S creen T alaris P ostScript gri
23. programmers were included Initial concepts were presented to the engineers on paper and the layout was refined based on their critiques When those providing input were largely satisfied the design was implemented in code Once the M files were finished they were distributed back to the engineers for evaluation The evaluation process was constantly in place for the duration of the code development phase of the project The final result was the desired series of command line functions that could successfully mimic their CIFER counterparts The various codes encompassed 49 functions and spanned 14 000 lines of code Extensive testing on the part of Ames engineers and the developer facilitated more robust code than might otherwise have been achieved One example of the success of the code was from one engineer running a series of CIFER cases in a few hours that might have otherwise taken two to three days to finish Suffice it to say that the engineers who regularly interact with CIFER were very enthusiastic about this new capability 21 3 1 2 Problems Encountered and Solutions The first and perhaps most fundamental problem encountered while developing this code was the structure of the interface itself Creating an interface is difficult because no one knows exactly what is desired until they see and interact with it Additionally no solution is necessarily the right or best solution On one hand there was a significant drive to have t
24. project application and experience It can display canned results to a wide variety of formats including PostScript X Talaris and others The textual interface was created in the 1980s in Curses format to run on Unix systems It has since been ported to run on a Unix emulation environment for Windows and on Linux An example of a CIFER screen is shown in Figure 1 4 The user would navigate through the screen using the arrow and function keys SOE OFTE VY LATSWP FRESPID 2 Comments Lateral frequency sweep for XV 15 in hover Controls AIL Outputs P RUD R AY Freq response output name XVLATSWP Cross correlate controls Y N Save results in file Y N Save results in d b Y N Generate plots Y N Figure 1 4 Example CIFER Screen Due to its long life and the constant work that goes into improving and updating CIFER it has many positive aspects that make it a premier frequency response analysis tool The algorithms have been proven to work providing quality results without program crashes Much effort has been given to creating robust methods of analyzing time history data and creating frequency responses CIFER includes many tools and utilities to assist in the analysis of frequency responses once they have been created such as frequency response arithmetic and bandwidth calculations The Curses interface is consistent and linearly driven which helps ensure that users enter data correctly Unfortunatel
25. s Figure 4 22 Aileron Roll Rate Responses Figure 4 23 Elevator Pitch Rate Responses RMS values for the time history data were manually calculated in MATLAB and compared to the equivalent values from CIFER calculations as based on the frequency responses These comparisons were conducted for both flight and simulation data Table 4 5 shows the results of the study The time RMS values were obtained from the pure control input data The two calculations compare well between the two domains Table 4 5 Frequency RMS compared to Time RMS Flight Simulation Freq Time Freq Time Ail 6 05 533 9 07 9 01 Ele 2 66 2 79 32 98 ZTT Rud 2 36 1 96 2 95 2 85 Whi Islo 1 05 Ltd 1 74 The crossover frequencies can also be examined by using the CIFER RMS calculations Flight and simulation results of the same on axis responses are shown in Table 4 6 All of the cutoff frequencies based on inputs agree well which means the flight control performance is consistent between the simulation and flight 51 100 deg wo oak Q ee o Co Phase I 3 Coherence Table 4 6 Cutoff Frequencies via RMS Flight Sim AIL 4 82 4 43 P 353 3 70 ELE 7 03 6 82 Q 5 62 4 87 RUD 6 21 6 03 R 2 53 3 89 WHL 6 04 5 59 R 2 53 3 83 Beta 85 none The differences in the rudder and wheel output calculations between simulation and flight are greater The phase slopes are different at t
26. the data content is poor Data is also considered poor or unreliable if the coherence is rapidly changing as illustrated in Figure 2 1 Poor data can result from noise gusts or off axis control activity CIFER algorithms use coherence weighting to determine the frequency ranges of a frequency response that have the most accurate data and only fit parametric models to these sections It is generally accepted that frequency ranges with coherence equal or greater than 0 6 are considered usable if they are not rapidly changing Reduced Accuracy x S iai 2 2 e 4 1 10 100 Frequency rad s Figure 2 1 Example Coherence Plot CIFER contains six primary programs to conduct analysis FRESPID MISOSA COMPOSITE NAVFIT DERIVID and VERIFY In addition there are three main analysis utilities that can calculate RMS values bandwidth properties and perform frequency response arithmetic The 11 functionality of these programs and utilities will be discussed in the following sections along with a brief description of the linear screen interface that drives them In addition to these tools there are a large number of other utilities for viewing results and organizing the CIFER database The primary CIFER programs this project dealt with were the first three main programs all three analysis utilities and a small selection of other programs 2 1 Creating a Frequency Response FRESPID FRESPID Frequency Re
27. CIFER MATLAB Interfaces Development and Application A Thesis Presented to the Faculty of California Polytechnic State University San Luis Obispo In Partial Fulfillment Of the Requirements for the Degree of Master of Science in Aerospace Engineering By Brian K Rupnik April 2005 Copyright 2005 Brian K Rupnik All Rights Reserved il Approval Page TITLE CIFER MATLAB Interfaces Development and Application AUTHOR Brian K Rupnik DATE SUBMITTED April 2005 Dr Daniel J Biezad AERO Advisor amp Committee Chair Dr Mark B Tischler NASA Army Committee Member Dr Eric Mehiel AERO Committee Member Dr Lanny V Griffin CENG Committee Member iii Abstract CIFER MATLAB Interfaces Development and Application Brian K Rupnik The Army NASA Rotorcraft Division Flight Controls Group Ames Research Center has developed and is maintaining a software package called CIFER or Comprehensive Identification from FrEquency Responses CIFER allows system identification in the frequency domain and is considered to be one the top resources for frequency analysis It provides methods to derive frequency responses transfer functions and state space models from a time sweep data The interface for CIFER was developed long enough ago that there is a significant demand for a modernization of the software To address the demand in the most complete manner would involve updating a very complex ser
28. CIFER cases typical of any detailed analysis The error checking associated with the data entry process for 19 CIFER inputs from the old interface which was split up over a long series of screens provided significant challenge The command line function had to emulate all of that error checking structure at one time Appendix A contains example screen shots of all the screens for the COMPOSITE program within CIFER Each screen is navigated through using the keyboard and error checking occurs when the user progresses from one screen to another 3 1 1 Development Process The first step in the development process was to determine the best method to get information from the CIFER database into the MATLAB workspace The solution lay in MATLAB s ability to create mex functions that can interface with Fortran or C code CIFER s original Fortran code is structured such that it has internal functions and subroutines to access important information By making the appropriate calls in the mex code these functions were successful in passing data from the CIFER Fortran into the MATLAB workspace and vice versa The mex functions became the building blocks of all the CIFER MATLAB interface code The concept of using internal CIFER functions with mex code was initially tested with some very basic functions designed purely to retrieve a few specific pieces of data from CIFER such as a frequency response name and description a
29. CIFER CONDUIT RIPTIDE Training Moffett Field Ames Research Center 2004 7 CONDUIT Version 4 1 User s Guide University of California Santa Cruz 41 071403 July 2003 8 Manur M H Dai W L Real time Interactive Prototype Technology Integration Development Environment RIPTIDE Installation and User s Guide UARC UC Santa Cruz Ames Research Center 9 United States Department of Defense Flying Qualities of Piloted Aircraft MIL STD 1797A 1990 10 Neal T P Smith R E An In Flight Investigation to Develop Control System Design Criteria for Fighter Airplanes Volumes 1 amp 2 AFFDL TR 70 74 Cornell Aeronautical Laboratory Inc Dec 1970 11 Mettler B Tischler M B Kanade T System Identification of Small Size Unmanned Helicopter Dynamics American Helicopter Society 1999 12 Mitchell D G Hoh R H Proposed Incorporation of Mission oriented Flying Qualities into MIL STD 1797A WL TR 94 3162 1994 References 1 Marchand P Graphics and GUIs with Matlab Boca Raton CRC Press 1996 2 McRuer D Irving A Dunstan G Aircraft Dynamics and Automatic Control New Jersey Princeton University Press 1973 61 Nelson R C Flight Stability and Automatic Control De ed New York McGraw Hill Inc 1998 Nise N S Control Systems Engineering New York John Wiley amp Sons Inc 2000 Nyhoff L R Leestma S C Fortran 90 for Engineers and Scienti
30. ER1 Figure 4 3 Mass Spring Damper Simulink Block Diagram The results of the simulation Figure 4 4 below were collected and formatted into a file readable by CIFER The time history then made the basis for new CIFER case that was created entirely in MATLAB The script that set up the case can be found in Appendix H 35 Phase deg Mag dB Coherence coco of BB o Input oO 10 20 30 40 50 60 70 80 90 Output 100 fi fi 1 1 fi 0 10 20 30 40 50 60 70 80 90 Time sec Figure 4 4 Mass Spring Damper System Input and Output The frequency response of the simulation is shown in Figure 4 5 below The natural frequency of the mode occurs is almost 5 rad s which corresponds to the calculated value of 4 83 rad s The case was also run from within CIFER for comparison to the MATLAB case the results of which are also shown in Figure 4 5 where the results overlay precisely In addition the CIFER results were compared to the results from using the MATLAB bode command on the transfer function as shown in Figure 4 6 There is a slight difference found at the mode and at high frequency which correlates to the drop in coherence at those regions CIFER er Matlab 90 CIFER 3 Matlab Bode oa a 90 ai fal x a ami 270 1 1 10 100 0 1 1 10 100 Frequency rad s Frequency rad s Figure 4 5 Matlab to CIFER Comparison Figure 4 6 CIFER to
31. ON OFF o 0 0 0 0 oo ww mow oo ww bow oOo paler See mow pl pt 1C vy 96 gt gt cifarith Structure fields are names cell array 2 of strings scalefac scalar array 2 1 spower scalar array 2 0 op EES v s td 14 outname string outid string cohopt N y isource D Rl J osource D Rl J minfreq scalar tof maxfreq scalar 0 nvalue positive scalar 0 unit A RAD Hz toscrn ON OFF gt gt cifplot Structure fields are array positive scalar array 5 0 correct integer array 5 o 1 names cell array 5 of strings gain scalar array 5 1 phase scalar array 5 0 spower scalar array 5 0 source f SD be eRe pminfrq positive scalar 0 pmaxfrq positive scalar bt pmin positive scalar array 5 0 pmax positive scalar array 5 0 pinc positive scalar array 5 0 device 1g ror ar ivi T grid ry N thick positive scalar 1 land port porge p xaxis positive scalar 0 yaxis positive scalar 0 Appendix G Online Help for Support Functions This Appendix contains the information that is displayed when the help functionname command is used in Matlab Function getfr Description This function allows the user to access the CIFER frequency response database and retrieve inform
32. RIVID constructs a state space system based on a series of related frequency responses It generates the appropriate control derivatives for the coefficient matrices of a state space system This is a very powerful ability as it enables the creation or validation of math models for 17 simulation and wind tunnel testing DERIVID works in a fashion similar to NAVFIT using a cost function to notify the user of the accuracy of the fit The final main program VERIFY allows users to validate a state space model found in DERIVID with new time history data in the time domain Typically the validation employs flight data taken from a doublet maneuver as opposed to a frequency sweep VERIFY will run the state space model using the time history inputs and compare the model outputs to the measured outputs This is an important step to provide confidence in the state space solution DERIVID and VERIFY use a combination of the screen interface and the prompt interface There are 17 and 18 screens respectively that accept user input and a lengthy series of prompts that guide the user through the calculations of the state space model Due to this complexity they were excluded from the scope of the current feasibility project 18 Chapter 3 Programming and Code Development The programming requirements set by the Army for this project included the development of a command line interface for three main CIFER programs FRESPID MISOSA and COMPOSI
33. S values for the aileron elevator and wheel inputs from the simulation are higher in magnitude than the flight RMS The likely explanation is that the simulation operators gave larger inputs to the control system than the flight operators to the flight test The rudder RMS values compare more closely between flight and simulation however simulation is still larger in magnitude The RMS comparisons of the pitch and yaw rates have the same trend as their respective inputs The values for the roll rate are almost the same which is not consistent to the difference in the magnitudes of the aileron inputs Figure 4 22 shows the comparison between the flight and simulation roll rate responses For additional comparison Figure 4 23 shows the same comparison in the pitch axis The results for the roll axis show a discrepancy in magnitude which could be due to the x axis moment of inertia estimated too large or the aileron control power derivative being estimated too small in the simulation Additionally the phase roll off is steeper at high frequency for the simulation which suggests there might be a time delay error The pitch response also shows an anomaly in the phase slope at high frequency which could be a time delay error as well 50 Mag dB deg 3 Phase Coherence I M y o Flight gi SE Fe el deg amp Phase M N Oo a 08 3 0 6 T 04 Q 1 10 100 O ia 7 i 10 Frequency rad s i i y Frequency rad
34. TE and an additional seven utilities RMS Bandwidth Frequency Response Arithmetic plotting case listing data storage and data retrieval These programs constitute a light version of CIFER that takes the user through all the frequency response conditioning and allows assorted manipulation and analysis of the responses The goal was to generate functional building blocks that had stand alone capability and upon which further development could take place In addition to the command line interface development of a GUI interface was also required The goal of the GUI development was to create a feasibility study to show how advances in GUIs could improve on the existing Curses interface Thus the development was focused on one main program COMPOSITE The primary concern was the layout and interface thus once one program was shown to work as a GUI others could be adapted fairly quickly This chapter discusses the process of development of both new interfaces for CIFER and the major problems encountered during the development process 3 1 Command Line Development A function in MATLAB accepts a list of inputs returns a list of outputs and typically is called from a single line often referred to as command line The key advantage of creating such an interface for CIFER is the ability to script multiple calls to the function The scripting ability greatly reduces the amount of time needed to create and run the large numbers of
35. The getfr function when given a frequency response name will return arrays for frequency magnitude phase and so on The other function writefr allows the user to change values in these arrays and pass them back in to be saved to the CIFER database The third function is called caselist and allows the user to query the database for which cases are stored for a particular program Due to the simplicity of these functions structures are not used to input data as is done for the other CIFER programs and utilities Thus these functions do not have the empty call feature that returns a list of structure field data types They do still include help documentation which can also be found in Appendix G getfr and writefr The getfr function is designed to give users access to all the arrays that make up a full frequency response The most complete call shown below will bring back arrays of all 10 frequency response fields The order in which these are returned is fixed therefore if only coherence were desired variables to hold information for frequency magnitude and phase would still be required as shown in the second function call which only returns four of the frequency response data fields gt gt name XVLATSWP COM ABCDE AIL P gt gt frgq mag pha coh gxx gyy gxy rel img err getfr name gt gt frgq mag pha coh getfr name Frequency responses can also be written back into the CIFER database through the w
36. UIT Difference Margin Frequency Margin Frequency Margin Frequency Margin Frequency Gain 1 106 67 0 41 10 67 0 41 10 3 0 48 3 57 13 18 Gain 2 22 84 13754 22 84 13 54 24 42 14 85 6 47 8 82 Phase 56 35 2 19 56 35 2 19 57 2 2 13 1 49 2 64 38 The differences between the two calculations are very small with the exception of the first gain margin frequency The difference in frequency could be a result of the time delay mentioned above however this would have more effect at higher frequencies It is reasonable that the results do not precisely match as they were obtained by different analytical methods Figure 4 9 contains a plot of the response with the gain and phase margins called out Gain Margin 1 Gain Margin 2 4 nh So o deg Phase Margin Phase Q3 So oOo 10 100 Frequency rad s Figure 4 9 Broken Loop Roll Gain Phase Margin Results In addition to the broken loop gain and phase margins CIFER can also determine bandwidth values from frequency response data Using both the MATLAB and Unix version of CIFER utility 8 for bandwidth calculations the roll attitude response to lateral stick was examined and compared to results obtained from CONDUIT These comparisons are shown in Table 4 2 and the difference between the two programs is very small There was no difference between the MATLAB and Unix runs of CIFER The ability to analyze nonparametri
37. _real 4 2000 ex_int 42 v gt gt myfun structure_example Figure 3 1 Name value Pairs and Structures The final interface for the command line code combined both name value pairs with the structure data type The structure was used in the background to track and store all the information from a CIFER case The functions accepted input as either a series of name value pairs or a structure Output had to be presented as a structure any other method would have produced too much clutter in the MATLAB workspace especially when multiple CIFER cases were considered The use of the structure helped streamline the internal error checking process Essentially for each field in the structure there was a series of checks run that mimicked the checks run in between each CIFER screen 3 1 2 1 Precision Errors During the earlier phases of development a small error caused by the transfer of data between CIFER and MATLAB was discovered The cause of this error was likely numeric precision discrepancies probably due to MATLAB using double precision compared to Fortran s usual single precision The result of these errors was that the same number stored in the CIFER database would return to MATLAB with variation in the ten or hundred thousandths decimal 23 place To date no major discrepancy has been observed in the overall results generated by CIFER Several tests were conducted to locate a more exact cause of the error The fi
38. able 4 3 Cutoff Frequency Results CIFER CIFER from Unix from MATLAB Channel Cutoff Freq Cutoff Freq LATA 7 00 7 00 PB 5 32 532 PHI 2 09 2 09 LATS 1 293 1 93 PB 3 75 3 75 PHI 0 57 0 57 Stick Actuator Cutoff Frequency Input Autospectrum gxx dB 0 1 1 10 100 Frequency rad s Figure 4 11 Stick and Actuator Input Autospectra The cutoff frequency for actuators can be indicative of the bandwidth frequency for some systems because the 135 degree crossover tends to occur during the phase shift that occurs at cutoff The roll rate tends to have more frequency content than the attitude as indicated by Table 4 3 which was why it was used for the previous bandwidth calculations and integrated to get the attitude response The cutoff frequency for roll attitude compares reasonably closely to the 45 degree phase margin bandwidth frequency for the lateral stick to roll attitude response 5 32 to 4 79 41 respectively Figure 4 12 shows the two frequencies marked on their respective magnitude and output autospectrum plots Output Autospectrum gyy dB nh Frequency rad s Figure 4 12 Cutoff Frequency Compared to Bandwidth Frequency 4 4 Analysis on Shadow 200 TUAV The following analysis was conducted on AAI Corporation s Shadow 200 TUAV for an Ames Research Center research project Shadow 200 has a wingspan of 12 75 feet length of 11 17 feet takeoff g
39. able or unstable that does not require assumptions of the system s structure u p 2 2 1 X f CIFER uses a version of the Fast Fourier Transform FFT known as the Chirp Z Transform CZT This transform removes many of the restrictions placed on the discrete Fourier Transforms and thus is very flexible as an algorithm Users have greater freedom to specify sample rates and resolution The algorithm only runs on a specified frequency range thus there are no wasted data points The CZT algorithm generates three important values that represent the energy of the system as a function of frequency input autospectrum G output autospectrum G and cross spectrum G The frequency response is then calculated using Equation 2 2 which is unbiased for output noise and biased for input noise H 2 2 G XX 10 One key feature of frequency response calculations is the coherence function which is a measure of data accuracy or content Coherence is determined by Equation 2 3 which will yield a value between 0 and 1 2 e Cle a 2 Vx 2 3 Coherence represents the energy of the system or the fraction of output power that is linearly related to input power When the system has high energy or excitation the coherence will be closer to 1 and the data content is considered to be more accurate If the system does not have enough excitation or is low energy the coherence will drop indicating that
40. andwidth and phase delay to determine handling qualities Phase delay is based on the twice 180 degree frequency as shown in Equation 4 13 Wise 180 4 13 T P 57 3 20 One concern with comparing the Shadow 200 data to the specifications as defined in these reports is that the studies were conducted for much larger piloted aircraft MIL STD 1797A uses AFFDL TR 70 74 as a basis for its bandwidth criteria This study was conducted to examine control system design criteria for fighter aircraft Thus the criteria from these documents will be scaled to Shadow as though it were a scaled down fighter Compared to the average fighter wingspan Shadow has a scale factor of about 3 Equations 4 14 through 4 16 show the form of the dynamic similarity laws that govern scaling where subscripts a and m denote actual and model respectively and N is scale factor 54 Roll Phase delay sec Length 4 14 i N i T Time Constant T 4 15 JN Frequency oO O JN 4 16 Figure 4 29 shows results for the roll axis plotted on the scaled WL TR 84 3162 handling specification for roll attitude The flight bandwidth is within the level 1 handling qualities boundary while the simulation result is level 2 Though the bandwidths of the two tests are similar the phase delay is significantly different 0 093 for simulation and 0 001 for flight This is a result of the differences in the phase curve of the frequency response sh
41. as a base to develop prototypes of these new interfaces as many companies already use it and the compiler addresses the concerns of those that do not CIFER is a large collection of programs and to develop command line functions to mimic the entirety of its capability is well beyond the scope of a thesis project It was agreed that functions to drive three of the major programs and several of the supporting analysis utilities would be a sufficient demonstration of a command line interface The goal is to create a function that can be called with a single line command that will perform all of the data input and error checking of its equivalent CIFER counterpart Development of the GUI was also limited in scope for the same reasons as for the command line functions The goal of the GUI is to show how a modern interface can enhance the functionality of CIFER Thus only one program will be used as a demonstrator In order to further tie the project into aerospace applications the code developed will be validated using real world problems In addition to checks verifying that results run in MATLAB are equivalent to those run from CIFER the MATLAB functions will be used to aid in a NASA research project Validations will include a simple mass spring damper system data from XV 15 tilt rotor aircraft flight tests UH 60 simulation data from a training course and finally a project involving handling qualities analysis of Shadow 200 a small reco
42. ase or run a case as a batch job CIFER CIFER Program MATLAB Function 1 FRESPID frespid 2 MISOSA misosa 3 COMPOSITE composite To facilitate the organization of information that goes into a CIFER case MATLAB structures are used to store case data The fields in these structures correlate to the field entries of the CIFER screen interface The MATLAB interface has optional inputs that allow users to specify values for a particular field using name value pairs This allows the user to modify parts of the structure without dealing with the entire structure Invoking the help functionname command from MATLAB will display examples of using each function as well as a detailed list of the CIFER information each field stores The contents of these help commands can also be found in Appendix C CIFER should be set to the database to be used by selecting the appropriate SIFDEF file prior to using these MATLAB functions If the database is changed by changing the SIFDEF file MATLAB must be restarted Features common to all three main programs The structures for each of the programs contain many fields thus the functions have been set up to do as much of the work of filling the fields in as possible for the user Each function can be called with no inputs or outputs to return a list of what type of data is contained in each field integers cell arrays scalar arrays etc The default value for each field is also specified in this
43. ater combined together using COMPOSITE which will be discussed below Of the three programs to be interfaced with MATLAB FRESPID is the most complex There are a total of nine screens in the Curses interface that accept user input and one of the screens can lead to two sub screens The user must specify information about the time history data build the controls and outputs using the measured data channels determine which frequency responses to calculate condition the data as desired specify the windowing information and set plotting and output options All of this functionality is supported by robust error checking that must be maintained in the command line interface Figure 2 2 shows an example of the FRESPID screen that controls the windowing of data Sc Window Parameters FRESPID 8 Final DT seconds 0 40000E 01 Window Id 20 chars TWIN In Out Decim Min Freq Max Freq points points Ratio radf sec rad sec SECOND WINDOW SECOND WINDOW 39 96000 999 1049 SECOND WINDOW 30 00000 750 274 SECOND WINDOW 20 00000 500 524 SECOND WINDOW 15 00000 375 137 Window size guidelines TWIN secs inputs DT As a function of individual time history events Enforced Window size lt or Smallest individual event Recommended Window size lt or 1 2 smallest individual event As a function of total linked time history duration all events Min recommended Window size lt or 1 3 linked record length Optimum cho
44. ation from the arrays stored there frq mag pha coh gxx gyy gxy rel img err getfr name Inputs name Name of Freqency Response string ex name XVLATSWP FRE A0000 AIL P Outputs Arrays of values lengths depend on database frq frequency mag magnitude pha phase coh coherence gxx JGXX gyy aGYy gxy gxy rel real img imaginary err error Function writefr Description The user is allowed to place frequency response data back into the CIFER database Minimum Input Requirements writefr name frq mag Maximum Input writefr name frq mag pha coh gxx gyy gxy rea ima err Inputs name Name of Frequency Response string ex name XVLATSWP FRE A0000 AIL P arrays to put into database frq frequency mag magnitude pha phase coh coherence gxXx GXX Syy gyy gxy gxy rea real ima imaginary err error NOTES There must be between 3 and 11 inputs Frequency 98 and magnitude must be included Other arrays can be included but they must be included in the order shown Function caselist Purpose Allow user to list present CIFER cases by program names caselist pgm Inputs pgm 1 2 3 4 5 6 7 8 9 10 TL 12 13 14 Outputs names program to check cases for FRESPID MISOSA COMPOSITE DERIVID model F matrix G matrix tau matrix H matrix M matrix VERIFY DERIVID results VERIFY results frequenc
45. c frequency responses is a useful feature of CIFER as it allows handling qualities of a system to be determined without identification of the full parametric math model 39 Table 4 2 Roll Bandwidth Results CIFER CIFER CONDUIT Diff From Unix From MATLAB 45 deg Phase Margin 4 79 4 79 4 77 0 42 6 db Gain Margin 6 89 6 89 7 05 2 27 The response for lateral input to roll rate output was used to generate the data from Table 4 2 First it was integrated to yield the attitude response and then the time delay of 0 0402 was applied The graph of this response with the bandwidths marked is shown in Figure 4 10 20 6dB Bandwidth Mag dB deg Fhase mM y oo S 100 Frequency rad s Figure 4 10 Roll Bandwidth for Lateral Stick Input to Roll Angle Response CIFER has an RMS utility that is capable of solving for the cutoff frequency of a response based on the energy content of the data Table 4 3 shows the cutoff frequencies solved for using both lateral actuator LATA and stick LATS inputs with roll attitude and rate responses by the Unix and MATLAB versions of CIFER The results show that the piloted input loses energy at a lower frequency and the actuator picks up content and operates at a higher frequency These results are confirmed by the input autospectra shown in Figure 4 11 with cutoff frequencies marked There is no difference in the results between Unix and MATLAB 40 T
46. call for the MATLAB interface It works nearly the same as the cifhq function The most basic call and the resulting output are shown below gt gt in cifxover gt gt in name XVLATSWP FRE A0000 AIL P gt gt cifxover in Search Range Min freq 0 139626 Max freq 31 415899 No OdB crossing found 180 n deg crossings for gain margin determination First 180 n crossover in selected range is Freq 27 818336 for 180 0000 deg Gain Margin 54 728527 dB First 180 n crossover in selected range is Freq 29 478342 for 180 0000 deg Gain Margin 63 286926 dB First 180 n crossover in selected range is Freq 30 105280 for 180 0000 deg Gain Margin 52 274651 dB The name value pair method of input works the same as described above for cifhg The call below shows an example of this used to run the utility with correction factors generate a plot of the output and save the new response The crossover characteristics portion of utility 8 is not tied 78 to either the linearized phase and coherence plot or the least squares fit in CIFER so those options are not available to the cifxover function gt gt gt gt gt gt gt gt gt gt gt gt name XVLATSWP FRE A0000 AIL P corrections 2 5 90 0 5 plot Y save Y savename XVLATSWP BAN A0000 AIL P cifxover 0 name name cor list corrections mpcplt plot save save Savename Savename toscrn off
47. caseout id crosscor id savdb id plot SCREEN 3 id evntnum id flghtnum id strttim id stoptim id source id thdt id biasflag id thfile SCREEN 4 id conchnl id conunit id conscfac SCREEN 5 id outchnl id outunit id outscfac SCREEN 6 id frall id frcalc SCREEN 7 id conditioning id condunit id savconth SCREEN 8 id winon id winid id winlen id wininpt id winoutpt id windec id minfft Control names user specified Output names Case name for freq responses may be different from the case name Cross correlate controls Y or N Save frequency response in the database Y or N Generate plots Y or N Event list i 1 10 Flight list corresponds to eventnum Start time for each time history Stop times Code for where time histories come from 1 for stand alone ASCII files 2 for TRENDS 3 for FLYTE 4 for READMIS gt 4 other sources as defined by source code Time interval between samples Bias and drift removal Y or N Array to store time history file names Primary channel names for each control Engineering units for control channels Scale factors for each channel Primary channel names for each output Engineering units for output channels Scale factors for each channel Compute all frequency responses Y or N Table indicating which responses to compute Blank if not to be computed to compute Array to store conditioning information The first row contains the cod
48. ceseiisaooineeiokh atechiade Rac pommmenele 3 1 1 2 Ames Research Center Planning Meeting cccccccscceesceseceseesecesecesecenecsecseecaeeeseesaeeneeeneenaees 5 1 2 PROJECTS COPE veeid 655535 B00 os a Pe SE ede Bo EGE eos is EE bs as teats EES 7 CHAPTER 2 SYSTEM IDENTIFICATION USING CIFER ccccsssccscssesccssssesscssssesssssssessssessesees 10 2 1 CREATING A FREQUENCY RESPONSE FRESPID wu0 ccccccecceccccceessssececececeessnssceceeeceesenseaeeeeeeeenas 12 2 2 MULTIPLE INPUT ANALYSIS MISOSA 0 0 cece ce cececececececececccececececececececececececececececececeeececececeeeeeeess 13 2 3 COMBINING WINDOWS COMPOSITE cccssccccccecsessssececececeenssnececececeeseaeeeseeeceesenssaeeeeeeeenes 14 2 4 ANALYSIS UTILITIES 330 55 c3 o fovescctectes es ceed cole oe ee Cee eae a cabot ees Deas 15 2 5 PARAMETRIC MODELING NAVFIT DERIVID VERIFY ccccccccccccccsesssceeececeesenseseeeeeeeees 17 CHAPTER 3 PROGRAMMING AND CODE DEVELOPMENT cccssssscsssssccsssssecesssssccesees 19 3 1 COMMAND LINE DEVELOPMENT cccccecececececececececececececececececececececececececececececececececececececececececess 19 ILI Development Process renet stess ductnauyinstg ites seat ies es ietansvab sea ieapect obuius eat tesecesthaieseative 20 3 1 2 Problems Encountered and Solutions ccccccccccsseecccecccccssssseccecccecsssssccsccccceccussaeseececessussneesesess 22 3 1 2 1 Precision Errors 3 1 2 2 Retaining Structure o
49. cture which is then modified for the specific response The full list of field names can be found in Appendix C gt gt in cifhq gt gt in name XVLATSWP FRE A0000 AIL P gt gt cifhq in Start Freq Start Mag Start Phase End Freq End Mag End Phase 0 1396 34 5106 2 6104 31 4159 39 9538 96 0240 Number of values in frequency response 923 k Handling Qualities Characteristics Start freq 0 1396 Start phase 2 6104 180 deg frequency 29 478342 Rad sec DB Gain 63 286926 dB Linear gain 0 000685 Hz 135 deg BW freq 29 463261 Rad sec DB Gain 61 210114 dB Linear gain 0 000870 Hz 6 dB Bandwidth frequency 29 437544 Rad Sec Another 6 dB Bandwidth frequency 27 796064 Rad sec Another 6 dB Bandwidth frequency 27 782501 Rad sec Another 6 dB Bandwidth frequency 24 950647 Rad sec Another 6 dB Bandwidth frequency 24 890440 Rad sec Another 6 dB Bandwidth frequency 17 924093 Rad sec Another 6 dB Bandwidth frequency 17 860493 Rad sec TWICE 180 FREQUENCY NOT FOUND All the information is displayed to the screen however output variables can be used to preserve these numbers There are two variables for storing cifhq output one for purely numeric values and a second that stores the words associated with each number Information about the frequency response can be saved to two additional variables Shown below is the function call with all outputs specified gt gt words
50. d lines Y or N line thickness 1 Default 2 double etc L andscape or P ortrait length of x axis length of y axis 95 Appendix F Structure Field Specifics for Analysis Utilities This Appendix contains the information displayed through an empty call to any of the functions Default settings are shown within a curly brace gt gt cifrms Structure fields are name source io spower minfreq maxfreq toscrn gt gt cifhq string T scalar 0 positive scalar positive scalar OFF ON Structure fields are name source minfreq cor List save savename mpcplt mpcmin mpcmax mpcdev lpcplt lpcmin lpcmax lpcdev lsfit lslow lsup lscoh lsdev toscrn gt gt cifxover string string N N Structure fields are name source minfreq maxfreq cor list save savename mpcplt mpcmin mpcmax mpcdev toscrn string D D p positive scalar scalar array 4 N ry N ryt positive scalar positive scalar 15 QO YT positive scalar positive scalar Ss roe TYI positive scalar positive scalar D Rh O N ry 1g Q ON OFF p positive scalar positive scalar scalar array 4 N ry string Nt ry positive scalar positive scalar 1g Q
51. d run batch job options name value pairs to set individual data fields optional e g casename XVLATSWP OUTPUTS out return structure for misosa data structure A template structure can be returned using an empty call gt gt out misosa EXAMPLE CALLS out misosa returns a mostly blank template misosa structure out misosa TEST 1 opens case named TEST returns misosa structure out misosa TEST 2 casename TEST2 winon Metro orrorro rr orily LA L L L saves TEST as TEST2 and the window selection is changed the structure is returned misosa TEST2 3 sends case TEST2 to batch nothing is returned NOTES A totally new case must be specified via a structure not by the desired new name i e misosa NEWNAME 2 will NOT save a blank case entitled NEWNAME into the database DESCRIPTION OF FIELDS IN INPUT STRUCTURE SCREEN 2 id casename Composite case name 85 id id id id id plot id id id id SCREEN id id SCREEN id id id id FUNCTION DESCRIPTION Function to open Function may be called with no arguments to return a template caseid casein caseout source savdb controls outputs winon 3 frall frealc 4 plotopt id id grid lrgplot plotdev pltminfrq pltmaxfrq Descriptive text for this case Frequency response prefix input Case part of freq resp name for output Source for input frequency re
52. dev output device Q MS C omprs V er S creen T alaris P ostScript in toscrn turn printed screen output ON or OFF Function cifarith Description This function allows the user to call CIFER utility 9 from the Matlab command screen This utility performs arithmetic operations on frequency responses The results are saved to a new frequency response out cifarith in options Inputs in input structure fields defined below Information on the details of the structure can be found with an empty call gt gt cifarith options name value pairs to set individual data fields optional esger asap Op Ges 93 Outputs out optional output allows user to set up a template structure using this call with no inputs gt gt out cifarith DESCRIPTION OF FIELDS IN INPUT STRUCTURE SCREEN 1 in names 2 input names in scalefac 2 scale factor in spower 2 power of s in op operation in outname Resultant Response Name in outid Resultant Response Description in cohopt N gt COH 1 0 Y gt COH COH of first response in isource Source of input responses D database F file in osource Destination for output D database F file SCREEN 2 in minfreq Minimum frequency to include 0 for default in maxfreq Maximum frequency to include 0 for default in nvalue Number of values in output response 0 for default in unit RAD or Hz 0 for default in toscrn Turn printed screen
53. ding rate response Table 4 7 135 Degree Bandwidth Frequencies Flight Sim PHI AIL 2 75 2 33 THETA ELE 5 86 4 27 PSI RUD 2 32 2 39 PSI WHL 2 62 3 78 BETA WHL 1 53 none These results compare reasonably well between flight and simulation As the cutoff frequency can be indicative of the bandwidth frequency it is useful to show comparisons between the two Figures 4 26 through 4 28 show these comparisons for each axis on the attitude response for the aircraft The results for the roll and pitch axes as well as the yaw flight data compare reasonably well The cutoff and bandwidth frequencies for the yaw simulation data do not compare as well which likely is a result of the differing output autospectra Flight Flight Sim o Bandwidth x Cutoff o Bandwidth x Cutoff Mag dB 90 45 g oO S A E1 a369 1 10 100 a8 1 10 100 Frequency rad s Frequency rad s Figure 4 26 Aileron to Roll Attitude Response Figure 4 27 Elevator to Pitch Attitude Response 53 Flight Sim o Bandwidth x Cutoff hase deg Frequency rad s Figure 4 28 Rudder to Yaw Attitude Response Bandwidth can be a useful indication of aircraft handling qualities and thus can be used to assist in control system optimization Classically bandwidth criteria for fixed wing aircraft are discussed in MIL STD 1797A and several subsequent studies These criteria use both b
54. dth ResultSet serei eesti cenieveecsiacasuiedeiti eatin leiden 40 Table 4 3 Cutoff Frequency Results cnra n n a i t i 41 Table 44 Full Range RMS Valesi occ 3 oaea acti ett E ea E R e i te 50 Table 4 5 Frequency RMS compared to Time RMS ssssseeesseersesereerrsresresrrsserrrsseesresresresressesreses 51 Table 4 6 Cutoff Frequencies via RMS eeii enes S SAE E Ea aTe 52 Table 4 7 135 Degree Bandwidth Frequencies esssseeesseeressseresrrsressesrrssesrrsstestesresrestressesresee 53 Nomenclature oe aS Z ZOK am o Longitudinal Acceleration Lateral Acceleration Vertical Acceleration Damping Coefficient Error Channel Feedback Channel Autospectrum Spring Coefficient Length Mass Scale Factor Roll Body Rate Pitch Body Rate Yaw Body Rate Frequency Domain Variable Longitudinal Velocity Longitudinal Body Acceleration Lateral Velocity Lateral Body Acceleration Vertical Velocity Vertical Body Acceleration Output Input Angle of Attack Sideslip Command Channel ee e ail dot ele rud whl XX xy yy Roll Attitude Phase Angle Coherence Pitch Attitude Time Constant Phase Delay 180 Degree Frequency Yaw Attitude Subscripts Actual Vehicle Aileron Time Derivative Elevator Model Vehicle Rudder Wheel Input Cross Output Chapter 1 Introduction The focus of this thesis is directed towards tools that aid in system identification System identification is the process
55. e The validations were central to the programming development as they allowed problems with interface structure to be fixed by the programmer as seen from the viewpoint of a user In addition to the validations conducted by the author discussed in this chapter engineers at NASA were supplied preliminary and completed versions of the interface to test in their own projects This allowed the engineers more input to the layout of the interface and provided additional testing for the new code 57 Chapter 5 Conclusions 5 1 Code Development The primary goal of this thesis was to create a modernized interface for CIFER that utilized both command line functions and a GUI interface within MATLAB Command line functionality was introduced for CIFER programs FRESPID MISOSA COMPOSITE the utilities for RMS Bandwidth and Frequency Response Arithmetic as well as additional plotting and data storage retrieval utilities This library of functions will provide a base from which further modernizations of CIFER can take place The command line functions proved to be complex enough that only an experienced user should employ them however the time savings to that user are significant In house users of CIFER at Ames Research Center were pleased with the new capability to script the set up and running of cases The development of the GUI as a feasibility study also proved successful the general look and feel of the CIFER screen interface was
56. e for the type of conditioning 1 for expansion 2 for decimation 3 for filter The second row stores the actual value If the filtering option is used this field stores the unit that the value is provided in If left blank Hertz is assumed Save files of conditioned time histories N no files A ASCII format only U unformatted only Y both types Entry for each window blank not to compute to compute Window descriptor Window Lengths Number of t h points input to the FFT Number of points returned by the FFT wininpt winoutpt power of 2 winoutpt must be le wininpt Output decimation factor for frequency responses Min FFT freq for this window 84 id maxfft Max FFT freq for this window SCREEN 9 id plotopt For each possible plot 0 not to plot 1 to plot id plotdev Device for plotting Q MS C omprs V er S creen T alaris P ostScript id grid Use grid Y or N id lrgplot Large plots Y or N id plotdec Decimate plots Y or N FUNCTION misosa DESCRIPTION Function to open save and run a MISOSA case Function may be called with no arguments to return a template mostly blank structure out misosa id cmd options INPUTS id either misosa structure or string with case name Information on the details of the structure can be found with an empty call gt gt misosa cmd command to have function perform a process 1 open MISOSA case 2 save a MISOSA case 3 save case an
57. e source and destination point to the database Any of these values can be modified either by changes to the appropriate field in the structure or by using the field name and value as a pair of inputs The above example could be compressed into one call as follows There are no output variables for this function and as before the toscrn field allows the user to turn off the printed return information gt gt cifarith 0 names XVLATSWP_FRE A0000 AIL P XVLATSWP_FRE AO0000 AIL R outname test outid arith resp from Matlab toscrn off cifplot The cifplot function is the MATLAB command line call for CIFER s utility 19 the frequency response plotting A basic set of calls to cifplot are shown below gt gt in cifplot gt gt in array 1 3 1 2 3 gt gt in names 1 XVLATSWP_FRE A0000 AIL P gt gt cifplot in Reading data Plotting successful After this call the standardized CIFER plot would be displayed to the screen As with the previous functions the call can be simplified using name value pairs as input The previous example could be reduced as follows There are no output variables for this function gt gt cifplot 0 array 1 2 3 0 0 names XVLATSWP FRE AOOOO AIL DD AM aN a Ss AECL Reading data Plotting successful 80 CIFER support utilities Three functions are available to facilitate the retrieval of data from the CIFER database
58. ed by similar checks within the GUI to make its use closely resemble the old CIFER interface Appendix A shows the COMPOSITE screens from both the old interface and the new GUI 3 2 2 Modern Updates to the Original Interface Perhaps the most visible change between the original interface and the new MATLAB GUI was the added navigational features CIFER was originally designed to run on VAX VMS systems and as a result uses the keyboard function keys for navigation There were menus that could be accessed for navigation within each program however they are not always intuitive in use Thus two primary navigation bars were added to the GUI interface The left bar found in Appendix A 29 Figures A6 through A10 offers users a visual description of each screen in the COMPOSITE program and allows users to access any of those screens with a mouse click The bottom menu displayed on the bottom of Figure 3 3 is a re creation of the original CIFER menu shown on top in Figure 3 3 and offers quick access to simple navigation and save or exit options Backup Update F Figure 3 3 Comparison of Navigation Menus Aside from the new navigation the largest change to the first screen Figure Al and Figure A6 is the addition of a browse feature This button opens the first window shown in Figure A11 and allows the users to browse the database for CIFER cases When a case is selected from the list its description is provided to aid in se
59. eeneecaaecsaecaecsaeenseenes 46 Figure 4 18 Vertical Velocity Perturbations eee esceessecssecsecsseceeceseesseeeeeeseaeeeaeesaaecaaessaeens 46 Figure 4 19 Lateral Velocity Perturbations Flight Data eee eeceeeeeeseeeneeeeecnseceaecnaeenseenes 48 Figure 4 20 Vertical Velocity Perturbations Flight Data eee eeeeeeeeeeeeeeeeseeeaeecnaeeneenes 49 Vili Figure 4 21 Comparisons with Exact 1 s Value oo e ccc ce seeesesecseeeesesscseesessessessssesseseessaeeaes 49 Figure 4 22 Aileron Roll Rate Responses sssini itni snieter E a eE Ea EEEE OEK SEa 51 Figure 4 23 Elevator Pitch Rate Responses sseeeseseessreesisrreserresresresrrssesrrssessresrestesressesresee 51 Figure 4 24 Rudder to Yaw Rate Response eeeescesseeesseceeceseeeseeeeeeseneeeseeeaeecaaecaecaecsaeenseeees 52 Figure 4 25 Rudder Output Autospectrum eee eesecesecsseceseceseeeseeeseeseneeeseecaeessaecaaeceeeeseeeseeeees 52 Figure 4 26 Aileron to Roll Attitude Response ee eeeeeeeeseeeseeeseecseecsseceaeceseceseeseeeseeeseeeneeeees 53 Figure 4 27 Elevator to Pitch Attitude Response ce esecescesceseeeseeeeeeeeseeeseesaaecaaecsaecnaeesaeeees 53 Figure 4 28 Rudder to Yaw Attitude Response eeeeeessecssesseceseeeseeeeeeeeseeeseecaaecaecaecnseesseenes 54 Figure 4 29 Scaled Roll Bandwidth Criteria 2 0 eee esseessecsseceseceseceseseeeeeseeeseessaecaecsaecsaeeeseenes 55 Figure 4 30 On Axis Roll Attitude Response eeeesceseceseceseceseeeeceseeeee
60. es There are three primary analysis utilities included in CIFER The first is the RMS program which is used to integrate the autospectrum over a desired frequency range to determine the mean square value of a signal In addition it can locate a frequency where the integrated RMS is a specific fraction of the full range value This is useful for determining properties such as cutoff frequency The second utility allows both the calculation of handling qualities and crossover characteristics The handling quality analysis looks for 180 degree 135 degree and 6dB bandwidth frequencies It provides values for the gain and phase margins at these frequencies Bandwidth is a useful indication of handling quality that can be identified from the nonparametric frequency response without first fitting a parametric model to the data The crossover calculation examines the magnitude of the response for sign changes and is useful for determining broken loop characteristics The bandwidth utility also provides some useful plotting features that will allow users to include a least squares fit over a portion of the phase curve and to solve for phase delay 15 The RMS and bandwidth utilities are driven by an interface different from that of the main programs described above Rather than using screens they use a linear prompt based interface Users are prompted for various data and results are then posted to the screen These programs were more challengi
61. esented using a Bode plot format that graphs magnitude on a log scale and phase of an input to output ratio against frequency Nonparametric models are useful in determining characteristics such as bandwidth time delay and pilot in the loop behavior They can also be used to validate math models and determine parametric model structure and order This project will be dealing primarily with frequency response analysis A parametric model assumes an order and structure with primary representations including transfer functions and state space models Transfer functions are pole zero representations of individual input output pairs State space models describe an entire system in terms of stability and control derivatives Parametric models are used primarily in control system design and for wind tunnel or math model validation 1 1 About CIFER Comprehensive Identification from FrEquency Responses There are many programs that offer time domain analysis of system response data but comparatively few that offer analysis in the frequency domain One very successful frequency domain program and the focus of this project is called CIFER or Comprehensive Identification from FrEquency Responses CIFER was developed by the Army NASA Rotorcraft Division at the Ames Research Center during from 1988 1994 and has been constantly updated modified and improved since It is used extensively by the Army NASA Rotorcraft Division and also by many commercia
62. f Code sii nescence avanti OS Wee Aiea Aenean Baa 26 3423 Complexity Of Se rna iE E EEE NAE E EEEE EEEE EEEE EAEE 27 3 2 GRAPHICAL USER INTERFACE DEVELOPMENT ccccccececececececececececececececececececececececececececececeeeeecs 28 B22 Development Process ensce nare EE EE E EEE EEE E EEEE EAEE 29 3 2 2 Modern Updates to the Original Interface ccccsccesssesscseescnseeecnecuseescnseeceesecseeseceeeeenaeeeeeaees 29 3 2 3 Problems Encountered and Solutions ccccccccccsseecececcccccssssccceccccccussecsccsccessuussesescecessuuanecceeees 31 CHAPTER 4 VALIDATION AND APPLICATION ccccssssscsssscccssscccssssscccsssscccsssscccssssscesseees 33 4 1 SAMPLE CIFER CASES iana a A A A A A R N E A 33 4 2 MASS SPRING DAMPER SYSTEM aeann e E E E TT 34 4 3 UH 60 SIMULA TION venn a A A RE E N E EEE ERE N a 37 vi 4 4 ANALYSIS ON SHADOW 200 TUAY ccccccccccccsesecesesesesesesesesesesesesesesesesesesesesesesesesesesesesesesesesesesens 42 441 Data Consistency CHECKS sioria acaba aE E ees A eee Sets 43 4 4 2 RMS and crossover comparisons cccccesccescese cess cee cnseesecusecasecaeeeseeeseeeseeceeesseeseceseceaecnaeenaeaas 49 4 5 SUMMARY OF VALIDATIONS j s 5sse5 ssestes cveabeas e oe a n oe a e Ee T E Ee ae ai 57 CHAPTER 5 CONCLUSIONG cssssssssssscsssssssssssssssssssssesesesesessssssssssssesesesesssessssnsnsssssseseseseseseres 58 5 1 CODE DEVELOPMENT viin fice oa e iE E A DEE REESE tive EE AEE ENEE Revues E REEE 58 5
63. f output pts Inputs AIL Outputs Windows Window z 000 size sec 000 000 000 000 Window size needed only for file input Figure A2 Original Interface Screen 2 63 a All Y N Clear COMPOSITE 3 AIL P R AY Figure A3 Original Interface Screen 3 SE e Plotting Options 1 to plot O not to plot COMPOSITE 4 Transfer fn magnitude Transfer fn phase Coherence Input auto spectrum Output auto spectrum Cross spectrum Error Heavy grid Y Large plot Y Plot output device P Ver C omprs Q MS T alaris P ostScript Figure A4 Original Interface Screen 4 You may B Submit a batch job and save case S Save case only E Exit without saving case Enter your choice E B Writing case info Preparing batch job The plot file will be u7 brupnik cifer jobs plots COM_XVLATSWP PSC 02 Job file u7 brupnik cifer jobs COM_XVLATSWP COM 05 submitted Strike anykey to continue Message from brupnik on ronin ttyq7 Wed Sep 6 13 39 52 ob 5238571 from batch file fu7 brupnik cifer jobs COM_XVLATSWP COM 05 has co pleted The log file is fu7 brupnik cifer jobs COM_XVLATSWP OUT 05 No error detected EOT gt Figure A5 Original Interface Final Screen New GUI Interface RREAEKKKKKKKKKKKKKKAKKKKKKKKKKKKKKKKKKKKKKERK Program COMPOSITE Multi Window Averaging x x x x x x x x k
64. for handling qualities output_n array containing numeric output from CIFER for handling qualities npts number of points in response FRinfo array containing information about the frequency response FRinfo sF sM sP sFm eF eM eP sPm the elements are starting and ending s e values for frequency magnitude and phase F M P m denotes value after modification by correction factors NOTE This function can be called with no outputs specified the results will output to the command window DESCRIPTION OF FIELDS IN INPUT STRUCTURE in name Name of Frequency Response string ex name XVLATSWP FRE A0000 AIL P in source Source F for file D for database in minfreq Minimum frequency for search range for handling qualities calculations Enter 0 for the default range in cor_ list Array with values as follows optional in cor_ list scor gcor ps td scor value for power of s correction gcor value for gain correction gt 0 ps value for phase shift td value for time delay NOTE an entry of 0 for any of these values is assumed to mean no correction shift etc 91 in save Y to save response N not to save in savename name to save response as optional ex XVLATSWP BAN A0000 AIL P NOTE If save Y and savename is left blank the program portion of the filename will automatically be changed to BAN in mpcplt create a magnitude phase coherence plot in mpcmin
65. for is in specifying the input case name The warning is much the same as for the output case name previously stated If the field is empty it will be filled in to match the case name Otherwise it retains the old entry Thus to fully change names from one case to another all three variables casename caseout and casein must be modified 73 CIFER analysis utilities The CIFER MATLAB interface includes three frequency response analysis utilities RMS Handling Qualities Stability Margin and response arithmetic Access to CIFER s handling qualities utility has been split into two parts handling qualities and crossover characteristics The current implementation allows full use of the various CIFER options for each utility from the MATLAB command line In addition to these analysis utilities the CIFER frequency response plotting utility may also be used from the MATLAB command line Below is a list of the utilities and their corresponding MATLAB calling function Appendix E contains the help files for each of these functions CIFER CIFER Utility MATLAB Function 7 RMS cifrms 8 Handling Qualities cifhq 8 Handling Qualities cifxover 9 Frequency Response Arithmetic cifarith 19 Plot Frequency Response cifplot All five utilities use MATLAB structures to organize the input variables and work almost the same as the three main programs They all use the same basic interface to convey information and run their re
66. ghfreq highest value in response rmsfrq frequency where RMS is desired fraction of full RMS value NOTES This function can run with no outputs specified the results will be displayed in the Matlab command window DESCRIPTION OF FIELDS IN INPUT STRUCTURE in name Name of Frequency Response string ex name XVLATSWP FRE A0000 AIL P in source Source F for file D for database in io I to integrate input auto O for output auto in spower value for power of s correction positive or negative integer 0 for none in minfreq start frequency for calculations 0 for default in maxfreq end frequency for calculations 0 for default NOTE set minfreq equal to maxfreq for a fraction of the full range RMS in toscrn turn printed screen output ON or OFF 90 Function cifhq Description This function calls the first half of CIFER s utility 8 the handling quality calculations output_s output_n npts FR_info cifhq in options Inputs in input structure fields defined below Information on the details of the structure can be found with an empty call gt gt cifhq options name value pairs to set individual data fields optional e g save Y Outputs output_s optional output allows user to set up a template structure using this call with no inputs and a single output gt gt out cifhq kkk OR output_s Cell array containing the resulting output from CIFER
67. he interface of new code mimic the old CIFER interface On the other hand it made equal sense to update older CIFER methods to more modern implementation and to make more drastic changes to the layout This question was more an issue for the GUI but still affected the command line interface One method that MATLAB uses to gather large amounts of input is through a series of name value pairs Essentially the name of a variable is given as input immediately followed by its value The pros are that the lists of variables do not have to be provided in a particular order and variables meant to retain default values need not be specified The con is that inputs are twice as long as they might otherwise need to be Figure 3 1 shows a simple example of name value input to a fictional function Another possible input method is to use a structure Structures are a parent data type in which various other data types can be stored Thus one structure might contain integers arrays and strings organized in the fields of the structure Structures are reasonably well organized and easy to deal with from a programming standpoint however the format can be intimidating to non programmers Figure 3 1 also shows a simple structure layout and how one structure used as input could replace name value pairs 22 gt gt myfun Namel Example Name2 4 2 Name3 42 gt gt structure examp le Structure examp le ex_string Example ex
68. he mode which suggests the simulation is more highly damped The outcome is surprising because the responses are very similar as shown in Figure 4 24 The primary difference is in the spike in the flight data around 2 rad s which is due to the poor coherence When the output autospectrum is plotted Figure 4 25 the difference between the flight and the simulation data is more pronounced This is the data that is integrated in the RMS calculation The plot of simulation data begins to roll off losing energy around 4 rad s which corresponds to the frequency from Table 4 6 The flight data begins to roll off just above 2 rad s also corresponding to Table 4 6 these autospectrum plots are the source of the discrepancy in the table Flight _RUD_R Sim_RUD_R Frequency rad s Figure 4 24 Rudder to Yaw Rate Response Output Autospectrum gyy dB Flight Sim J aoaaa n tnt ra Parara 1 10 100 Frequency rad s Figure 4 25 Rudder Output Autospectrum 52 CIFER offers a utility for analyzing bandwidth and crossover properties of frequency responses Use of the handling qualities portion of the Bandwidth utility allows identification of 180 degree and 135 degree bandwidth frequencies and gains Table 4 7 shows the 135 degree bandwidth frequencies for the main on axis responses in flight and simulation The attitude bandwidths were determined by integrating the correspon
69. htforward as opening a case The three calls below will execute the batch job for the appropriate program FRESPID MISOSA or COMPOSITE for the XVLATSWP case The batch call will pause MATLAB until the batch job has finished display information about the job and log files and report if there are potential errors with the batch job Only the screen output for the frespid call is shown to conserve space 71 gt gt frespid XVLATSWP 3 Job file usr cifer jobs FRE_XVLATSWP COM 01 submitted Waiting for batch to finish Batch completed The log file is usr cifer jobs FRE XVLATSWP OUT 01 No error detected gt gt misosa XVLATSWP 3 gt gt composite XVLATSWP 3 The example below will take the XVLATSWP case copy it to TEST and then run it as a batch job all with a single command The batch command always saves the case before it sends it to the batch job To verify the name change was correct the structure which will contain information for TEST is returned gt gt out_f frespid XVLATSWP 3 casename TEST caseout TEST frespid This section contains information specific to the frespid function and will primarily cover the mechanics that mimic the functionality of FRESPID screens 7 and 8 conditioning and windowing of data Data conditioning in screen 7 is handled with two fields in the frespid structure The first field conditioning contains two rows the first row stores a flag
70. ice Window size lt or 1 5 linked record length Min ID freq for each window 2 pi TWIN Max ID freq 2 pi 5 DT The inputs plus outputs must equal a power Figure 2 2 Example FRESPID Screen 2 2 Multiple Input Analysis MISOSA Engineering problems involving flight vehicles tend to involve systems with multiple controls During flight tests frequency sweeps are performed for a single axis However there may be 13 secondary off axis control inputs This coupling has significant potential to distort the response identification of the system CIFER addresses this issue with a program called MISOSA or Multi Input Single Output Spectral Analysis MISOSA analysis on the user level is very straightforward The primary control of interest is specified along with any other controls that are to be considered secondary The program then performs spectral analysis to remove the effects of the secondary controls from the primary control response It uses a matrix inversion of the input autospectrum at each frequency point to accomplish this This general form is shown in Equation 2 4 G is the matrix of auto and cross spectra for the inputs and G is a vector containing the cross spectra for each control input and the single output in question T f G f G f 2 4 There are much fewer screens that drive MISOSA compared to FRESPID The primary information needed is details about where the frequency responses are
71. ics on this step and all following steps will be discussed with more detail in Chapter 2 The data is divided into time windows which allow the algorithms to accurately extract both high and low frequency content from the data The frequency responses are then calculated for each of the desired combination of inputs and outputs for each time window 10 Rudder Deflection deg Oo 10 20 30 40 50 60 Time s Figure 1 2 Frequency Sweep Example UAV Flight Data In the case of a multiple input system the frequency responses can be conditioned to remove the effects of correlation between different inputs After this conditioning the frequency responses from each separate window are combined to form a single frequency response based on the most accurate segments of each window The combination is achieved by optimizing the data that has useful content in both low and high frequency regions This is the last step in the frequency response generation process Less involved analysis compared to full state space model generation can include calculations of RMS cutoff frequencies bandwidth crossover characteristics and data consistency checks using frequency response arithmetic All of these tools can give useful insight to system behavior without generating more complex math models CIFER offers more complete identification through programs that will fit transfer functions to individual responses and full state space models to a se
72. ies of programs with modern graphical and command line interfaces This project is beyond the scope of an Aerospace Master s thesis However before the Army devotes resources to this task they desire a proof of concept This thesis is that proof of concept Many users of CIFER agree that having CIFER programs and utilities usable from the MATLAB command line or modernized graphical interface would be a major benefit The Army agreed that development of a CIFER MATLAB interface would be both a useful tool and a stepping stone for where they would like to take CIFER in the future There are two main tasks that make up this thesis The first task is the development of a CIFER MATLAB interface both at the command line and in a graphical user interface This interface covers some but not all of the programs in CIFER enough to show that the interface works and makes use of CIFER more efficient The second task is to validate the new interface through a series of projects including analysis of a modern Unmanned Aerial Vehicle UAV Both tasks were successful in the eyes of the Army sponsors and ongoing work is being conducted to implement the work from this thesis into the whole of the CIFER program suite Acknowledgements The author would like to recognize and thank Dr Daniel J Biezad Department Chair at Cal Poly San Luis Obispo CA and Dr Mark B Tischler Flight Control Group Leader Army NASA Rotorcraft Divis
73. information about the frequency response optional FRrng minFreq maxFreq NOTE This function may be called with no outputs specified 92 the results will be displayed in the command window DESCRIPTION OF FIELDS IN INPUT STRUCTURE in name Name of Frequency Response string ex name XVLATSWP FRE A0000 AIL P in source F for file D for database in minfreq Minimum frequency for search range for crossover calculations Enter 0 for the default range in maxfreq Maximum frequency for search range for crossover calculations Enter 0 for the default range NOTE Set minfreq equal to maxfreq to search for a desired fraction of full range RMS in cor list Array with values as follows optional cor list scor gcor ps td scor value for power of s correction gcor value for gain correction gt 0 ps value for phase shift td value for time delay NOTE an entry of 0 for any of these values is assumed to mean no correction shift etc Leaving this input out will default the entries to 0 in save Y to save response N not to save in savename name to save response as optional ex XVLATSWP BAN A0000 AIL P NOTE If save Y and savename is left blank the program portion of the filename will automatically be changed to BAN in mpcplt create a magnitude phase coherence plot in mpcmin minimum frequency for mpc plot in mpcmax maximum frequency for mpc plot in mpc
74. ints are set to zero while a window is turned on they will be automatically calculated using the same logic that CIFER employs If only window lengths are specified and all other related fields set to zero then the function will generate the same values that CIFER would if the screen were stepped through manually Thus the following call would achieve the same result as entering the window lengths in CIFER screen 8 and pressing the PF1 key until CIFER finished filling in the fields gt gt in frespid NOWINS 1 gt gt frespid in 2 name NEWWINS winlen 45 40 30 20 15 The example retrieves a theoretical template case that has no window data entered and copies it to a case that now has fully specified window data Users can specify their own values to fields such as maximum frequency and the automatic calculations will not occur unless the specified values violate the rules governing the variables For instance if the numbers of input and output points do not add to a power of two the function will adjust them and issue a warning Care should be used when using one case as a template for another and modifying the window lengths to make sure that old values are not written into the new case misosa and composite The CIFER programs MISOSA and COMPOSITE are less complicated than FRESPID Accordingly the corresponding MATLAB functions have far fewer field entries and automatic corrections The main issue to be alert
75. ion Ames Research Center CA for their invaluable support guidance and encouragement throughout this project Dexter Hermstad was instrumental in providing knowledge and input throughout the programming development process and deserves many thanks The efforts of Dr Colin Theodore Dr Jeff Lusardi Kenny Cheung Chad Frost and Rendy Cheng on behalf of this endeavor are also greatly appreciated Everyone in the Flight Control Group was highly supportive of the effort and they all deserve recognition for their assistance at various stages of this thesis Table of Contents TABLE OFFIGURES iecisssssccscssescosecscsessensesvsscessvesessssvesesessvecessessesuissusatesdaseesessstesdussesessstvesessess oseveteosee Vill LIST OF TABLES weiscccvssscossseciecesvecsesesvecsecnsvevseonssesieveasessessasesssscedestescodscteceobestesssbestesssdectoossiecdesessessesusvecsess Ix NOMENCLATURE cciscscossssccsteevecsesesvensecnssesseonssesietenssctessessssccossvensessesvasecsnscodsviscessesdassosdent cssbestessossvesseeses X NOMENCLATURE 3 oiiiciscesesssccssscdececstusssscsacsletesenssscsvssteossossessedsocsseusnsotasaccsssavssnssuasedssessesesessssceaunesoeveesccnesonse X CHAPTER 1 INTRODUCTION scsssisiisscssssscescstsossssostssesessseuscessosstssasoncssacsbecisossdecssssteasossvsvasssesssestsoesses 1 1 1 ABOUT CIFER COMPREHENSIVE IDENTIFICATION FROM FREQUENCY RESPONSES c00000 2 1 1 1 What CHEER Encompasses susan duces d sxeevuan ane etpiaca
76. is is largely the subject matter of the CIFER documentation AIL_P aE FER Matlab AIL_R 20 0 T T ae ei 60 100 _ 180 a D az D a 3 p 3 A o on P d f 4 nel a 360 4 2 AIL_AY AIL_VDOT Mag dB 5 o 8 Mag dB Bio 8 Q9 ap m co oO 0 Phase deg D O Phase deg 180 lulh 1 10 100 0 1 1 10 1 Frequency rad s Frequency rad s D oo Figure 4 1 XVLATSWP Validation Examples Mass Spring Damper System A second elementary check on the MATLAB interface was conducted using a single input single output SISO mass spring damper system as an example The primary goal for the example was to provide students with a simple system to analyze using CIFER The system used is shown in Figure 4 2 and its transfer function was solved from the governing differential equations in the form of Equation 4 1 OUT s k SS e 4 1 IN s Ms bs k 34 inp y x t Figure 4 2 SISO Mass Spring Damper System The system was modeled in Simulink by attaching a custom chirp signal generator to the transfer function as shown in Figure 4 3 The chirp generator was written to provide a frequency sweep as input The spring constant K was set to 7 and variables M and b set to 0 3 and 0 1 respectively These values correspond to a natural frequency of 4 83 rad s and a damping of 0 035 Response Transfer Fen Terminator2 chirp generator To_CIFER To_CIF
77. l aerospace companies Some of the applications at the Ames Research Center have included development of control laws for the UH 60 identification of the XV 15 tilt rotor demonstrator assistance in CH 47 control development investigation of slung load dynamics and a wide variety of work involving unmanned aerial vehicles UAVs CIFER is considered to be one of the best programs available for frequency domain analysis 1 1 1 What CIFER Encompasses CIFER contains all the programs and tools necessary to convert time history data into frequency responses and use those responses to identify the system in question This analysis includes the identification of single transfer functions entire state space systems and various properties of responses such as bandwidth and crossover characteristics Systems are not limited to flight vehicles and can be as simple as a single input single output mass spring damper setup or as complex as a multi input multi output rotorcraft model The basic flow of the program begins with time history data The data is generated using a frequency sweep maneuver in flight or simulation to excite the system over a wide range of frequencies Frequency sweeps are generally characterized by a sinusoidal motion with a constant increase in frequency preceded and followed by a period of steady state flight as shown in Figure 1 2 The first step in CIFER is to transform this time data into frequency responses More specif
78. lank frespid structure f An Erespid thename XVLATSP2 Fill in all the necessary information to make the case in casename thename in comments Matlab created XVLATSWP case new functions in caseout thename f in crosscor PY f in plot IN Time history selection parameters f in source 1 fin evntnum 1 2 883 884 f in flghtnum 1 2 150 150 f in thdt 0 004 channel definition parameters Tolnecont role 1 2 AIL RUD f_in conunit 1 2 deg deg f_in outputs 1 4 P R AY VDOT f in outunit 1 4 deg s deg s ft sec2 ft sec2 in conchnl 1 2 1 D645 D284 f in outchnl 1 4 1 V012 V014 A300 VDOT f in outscfac 1 3 1 0 0175 0 0175 32 1740 Conditioning parameters f in conditioning 1 1 2 3 2 gt f_in conditioning 2 1 2 4 25 Frequency response selection parameters f_in frcalc 1 4 1 2 Window Parameters 45 SECOND WINDOW 40 SECOND WINDOW 30 SECOND WINDOW 20 SECOND WINDOW 15 SECOND WINDOW f in winlen 45 40 30 20 15 f in winon 1 5 Save the structure into the database respid fin 2 in winid 0 oe Set up blank misosa case m_in misosa oe Fill in appropriate values m_in casename thename m_in comments Matlab created XVLATSWP case m_in casein thename m_in caseout thename m_in controls 1 2
79. lecting the desired case The case name can then be loaded into the text field on screen one Screen 2 Figure A2 and Figure A7 originally completely consisted of text field entries Many of these fields corresponded to data that could only be one of two choices yes or no for example The MATLAB GUI uses toggle buttons to simplify this process and reduce the amount of error checking necessary Additional browsing capacity has been added to help data entry inputs and outputs can be selected from a list of all inputs or outputs in that case For example a user might load a case into COMPOSITE and not remember which inputs and outputs were used in previous FRESPID cases The input and output buttons Figure A7 call up a new screen the second window in Figure A11 that displays all the inputs or outputs that exist in a particular case by querying previous MISOSA or FRESPID cases The desired inputs or outputs can be selected and then loaded back into screen 2 30 The third screen is used to match which input and output pairs the program will operate on Originally these pairs were selected through an asterisk as shown in Figure A3 The new interface Figure A8 is toggle button based and more features have been added to allow an entire row or column to be selected by clicking on the heading button Similarly toggle buttons were used for screen 4 as shown in Figure A4 and Figure A9 The last screen in the original code offer
80. line of MATLAB This enables easy retrieval of information in CIFER to the MATLAB workspace for analysis and allows the scripting of multiple batch jobs The command line functions will perform all the operations and checks contained in each screen of a CIFER program at one time While this allows the user great flexibility the linear setup of the CIFER screens is bypassed Thus new users are encouraged to verify the data entered in the command line by running the corresponding program within CIFER and checking each screen Contained within this document are a series of short examples detailing the various features and layout of the command line functions These examples are supplemented by the online help information accessed using the command help functionname in MATLAB and two extended examples all of which are contained in this document s appendices 68 CIFER main programs The three main CIFER MATLAB interface functions are designed to give users the functionality of the CIFER programs FRESPID MISOSA and COMPOSITE by making calls from the MATLAB workspace All the functions are structured in the same manner with two required inputs The first input is either the name of the desired CIFER case or a MATLAB structure that contains information representing a CIFER case The second input is the flag to execute a particular command on that case The commands that can be performed are to open a case save data to a c
81. mper system was modeled as a potential example to new users of CIFER Once the codes were more developed more complicated validations were employed The first was a closed loop investigation of a UH 60 simulation and the second was a series of data consistency checks and cutoff frequency analysis on the Shadow 200 Tactical Unmanned Aerial Vehicle TUAV 4 1 Sample CIFER Cases CIFER installations come with a series of time histories for the XV 15 and its documentation uses related sample cases as examples New users of CIFER typically learn the interface by working through the set up of these cases Thus the cases made excellent initial comparisons to determine if the MATLAB interface was working correctly At this stage the analysis was essentially creating plots of cases set up and run from CIFER and cases set up and run from MATLAB The only major problem found was the numeric precision error discussed in Chapter 3 Examples of the scripts used to set up one example case can be found in Appendix I A series of example comparisons for the lateral sweep are shown in Figure 4 1 The aileron input is plotted against roll rate yaw rate lateral acceleration and vertical velocity perturbation The important aspect of the figure is that the plots comparing the results set up and run from MATLAB to those set up and run from CIFER match Additional plots and analysis for the XV 33 15 cases have not been included as the analys
82. mplate structure The call to create a template structure is very simple as shown below After the calls each of the out_x variables contains the basic skeleton used to create CIFER cases This feature is useful to ensure that the field names and data types are correctly specified and the various arrays are the correct length The fields are all initialized to the same defaults as found in a new case of the appropriate CIFER program with the exception of plotting which defaults off for all three gt gt out_f frespid gt gt out_m misosa gt gt out_c composite Opening Cases 1 as second input Retrieving case information from the CIFER database is accomplished by calling the MATLAB functions with the case name specified as the first input and 1 for the second input as shown below All of these calls would write the information from the CIFER case XVLATSWP to the appropriate MATLAB structure out_x out_x can then be viewed and modified as needed 70 gt gt out_f gt gt out_m gt gt out_c frespid XVLATSWP 1 misosa XVLATSWP 1 composite XVLATSWP 1 Saving Cases 2 as second input Saving a case works much the same as opening cases A simple call to save a case might involve specifying an existing case to be renamed as a new case This call is shown below the existing XVLATSWP case is opened renamed to TEST and then saved back into the database as a new case
83. name cor list corrections toscrn off This call utilized the name value pair style of input The name value pairs do not have to be specified in a particular order as long as they occur in pairs Note that the flag of O for the first variable makes the function automatically use the default structure set up in cifhq If the 0 flag is used any response specific inputs must be specified or the call will result in errors The template does not assume default input or output response names TT The cifhq function supports the utility 8 ability to create plots and perform a least squares fit on the data There are a number of options for turning on the plots and setting their ranges The call below implements the full series of plotting available from utility 8 gt gt name XVLATSWP FRE AOOOO AIL P gt gt plot TY gt gt linplot Y gt gt leastsquare Y gt gt cifhq 0 name name mpcplt plot lpcplt linplot lsfit leastsquare tosern off This call would display a total of three CIFER plots to the computer screen The values for the ranges on all the plots are initialized to a default value which can be modified by changes to the appropriate fields see Appendix E in the structure cifxover The second half of the handling qualities utility in CIFER allows the user to find crossover properties for a frequency response This ability has been split out into a separate function
84. nd display them in MATLAB Modifications were made to the data in MATLAB and fed back into CIFER Accessing the frequency response in CIFER verified whether or not the change was successful When data is first processed in CIFER it becomes known as a case which encompasses a series of measurements from a system For example one of the sample cases included with CIFER is called XVLATSWP which is for lateral responses from the XV 15 research vehicle in hover The case structure has all the information that dictates where data is located how to 20 interpret it how to condition it and how to window it It is this information that must be passed into MATLAB in order to facilitate running a case from a command line interface The mex building blocks were expanded upon to read and write the full gamut of information CIFER uses to create and run its cases The problem with these building blocks is that they have many required inputs or outputs depending on read or write and no inherent error checking This made use of the initial code blocks themselves very difficult It was necessary to create higher level MATLAB M files that would call the mex building blocks These functions were designed with the full range of error checking found in the CIFER screens and could format input data to be fed into the mex code blocks Designing the final interface for the M files was an iterative process in which several Ames engineers and
85. ng to capture in a command line format because the inputs had to be entered at one time and the outputs displayed at one time Figure 2 3 shows an example of an RMS calculation and the prompts that drive it 4 Program RMS determines mean square value of signal by integrating autospectrum over desired freq range Enter source of data Ff ile or Df atabase D Enter response name A60 or lt CR gt to quit Input XVLATSWP_COM_ABCDE_AIL_P Integrate input auto I or output auto 0 Enter I or O I Do you want to correct the spectrum with a power of s Y N Y Enter a positive or negative power 1 NPTS 981 First and last frequencies rad sec OMEGA 1 0 1396262 OMEGA 981 31 41590 Enter starting and ending frequencies for integration in rad sec Enter OMEGA1 and OMEGA2 for either default lt CR gt for entire range or single value desired fraction of full range RMS Input Integrating data for mean square value Results Mean Square Value dae Root Mean Squared Value 3 434087 Do you want to enter frequency range Y N Figure 2 3 Example RMS Prompts The last function performs frequency response arithmetic Responses can be modified individually using scale factors and powers of s and then combined with basic arithmetic operations The utility will perform these operations for magnitude phase coherence and the auto and cross spectra as desired The option to perform this a
86. nnaissance UAV all of which are shown in Figure 1 5 and Figure 1 6 inp x t Figure 1 5 Mass Spring Damper Left XV 15 Right Figure 1 6 NASA Sikorsky UH 60 RASCAL Left Shadow 200 TUAV Right A final important aspect of the project is that the code must be developed and structured such that developers at Ames Research Center can continue the work beyond the scope of this project with the final goal of distributing it to CIFER users This translates into creating the appropriate documentation and code structure to allow this project to be integrated into the existing CIFER code structure efficiently Chapter 2 System Identification using CIFER This chapter primarily contains information found in the CIFER user s manual and is intended to give readers a solid understanding of how CIFER generates a frequency response This will help provide insight into the analysis of the validations described in Chapter 4 CIFER produces both nonparametric and parametric models for systems in the frequency domain The nonparametric frequency response should be considered the core of this thesis however parametric modeling is still very relevant to CIFER as a whole A frequency response is a complex valued function that relates the Fourier Transform of system output to system input The general form of a frequency response is shown in Equation 2 1 The frequency response is a full description of system dynamics st
87. ons however to provide CIFER the information without using the screen interface The MATLAB structure that holds the case information is set up to collect the same information that the user would normally provide through the screens All of this functionality was accomplished without modifying any original CIFER code thus no provisions will be necessary in order to maintain that original code when the new interface is added to the full CIFER software package Other utilities such as the RMS plotting and arithmetic calculations have the screen interface embedded within the functions that run the calculations For the MATLAB interface to work the original CIFER functions had to be altered to disable the screen interface These alterations are well marked with comments as CIFER lead programmers will ultimately need to incorporate new methods within existing code to accommodate the changes and prevent the need to maintain two separate source files The utilities that use the command line interface RMS plotting and handling qualities were dealt with much the same as the screen based utilities due to the command line prompts being embedded within the code for calculations 3 1 3 Complexity of Use A side effect of the command line interface was that it required users to be reasonably familiar with CIFER The interface is not as intuitive as a graphical interface because everything happens at once There are no screens with helpful n
88. otes and error checking to step through If a case has 27 errors they are all displayed at once when the program finishes Thus a new CIFER user could get more lost and confused by starting off with the command line interface The power of the interface as mentioned before is the ability to script multiple cases which is something beginning users would find less beneficial than experienced users A detailed help document was created for the command line interface to help users understand how to employ it The text of this document can be found in Appendix B and Appendices C through I contain the appendices from the original document Ames engineers were provided copies of the document along with the software when they employed the new interface in their projects The document was written with the assumption that a reader would already be familiar with CIFER 3 2 Graphical User Interface Development COMPOSITE was selected as the test program around which to develop a MATLAB GUI It was deemed sufficiently complex to make a useful example and it is one of the more commonly used CIFER programs from those selected for this project The primary goal of the GUI development was to show that first it could be done and second the process of data entry could be enhanced using more modern GUI tools An important secondary goal was to create the GUI as a rough template that could be efficiently applied to the rest of the CIFER programs in
89. own in Figure 4 30 with the 180 degree and twice 180 degree frequencies marked The simulation data rolls off much more quickly than the flight data The discrepancy between the two tests would need to be cleared up before a solid conclusion on the handling qualities could be reached 20 30 a D 00 90 135 D Bg 2 360 0 5 1 1 5 2 2 5 3 3 5 4 0 1 1 10 Roll Attitude Bandwidth rad sec Frequency rad s Figure 4 29 Scaled Roll Bandwidth Criteria Figure 4 30 On Axis Roll Attitude Response The analogous scaled specification for pitch attitude bandwidth is shown in Figure 4 31 Only flight data was calculated as the simulation data did not exhibit a 180 degree phase crossing at 55 Flight Simulation 100 Phase delay sec o 2 A o nm Oo 5 ad gt S a D o s bad mM high frequency as shown on Figure 4 32 Flight data with marks for the 180 degree and twice 180 degree frequencies is also shown This handling specification is for Category C flight which covers terminal flight phases that typically involve non aggressive maneuvers and accurate flight path control such as takeoff approach and landing The results predict level 1 handling qualities The phase delay 0 009 is fairly small As the calculations had to be conducted on a low coherence portion of the phase plot it is probable that the phase delay is higher though cleaner data
90. passed by a single command line call combined functions could still allow experienced users to efficiently script setting up cases Both of these programs would benefit greatly from updated GUI interfaces which could easily run multiple command line functions behind the screens In addition to finishing development on the remaining CIFER functionality the code will need to be production tested to assure that it is robust Engineers at Ames Research Center are making the first steps towards this process by applying the current code to larger more involved tasks Once the code is fully developed and its capacity verified to the satisfaction of industry users the new code can be released as a standard feature of the CIFER suite The work accomplished in this thesis goes far towards realizing the possibilities of the new interfaces as a reality that will greatly benefit all users of CIFER 60 Bibliography Works Cited 1 Hamel P G Editor Rotorcraft System Identification AGARD AR 280 1991 2 The Mathworks Inc www mathworks com 2004 3 Thompson K D Ames Imaging Library Server ails arc nasa gov 4 AAI Corporation Shadow TUAV www shadowtuav com Accessed 2004 5 Tischler M B CIFER version 2 1 Comprehensive Identification from Frequency Responses Vol 1 Class Notes Vol 2 User s Manual Army TR 94 A 017 8 NASA CP 10149 10150 September 1993 6 Tischler M B Mansur M H Combined
91. preserved while adding various elements that enhanced and accelerated the case set up process over the previous Curses interface The introduction of modern navigation tools such as browsers and menus will afford users more awareness of where they are in the set up process and improve the learning curve of new users The last important aspect of the programming was the attention to the fact that the code will be built upon by other programmers at NASA Thus the code was well commented and documented The general structure of the code was made to be as uniform as possible While the author s development of this code has ceased programmers at Ames are continuing development and report that the efforts made to create an easily modifiable code were successful GUIs for both 58 FRESPID and MISOSA have been created with a minimum hassle Much of this success is due to the reviews of in house users and programmers 5 2 Analysis The most important conclusion from the analysis was that the code developed for this thesis worked Numerous comparisons between cases run in CIFER and cases run in MATLAB were made and no appreciable difference was found Slight discrepancies were discovered but careful examination of these suggested that they were caused by machine precision issues probably because MATLAB automatically makes all numeric variables double precision The differences due to precision error were both negligible and tended to occur in por
92. ries of responses In addition to these powerful analysis tools CIFER offers utilities for plotting and organizing output as well as managing the storage and organization of data One last important tool is the ability to validate state space models against other time history data These validation time histories are typically generated using a doublet maneuver as opposed to a sweep A doublet is a short maneuver that moves through a range of motion for a control surface beginning and ending in steady level flight as shown in Figure 1 3 Rudder Deflection deg 0 5 10 15 Time s Figure 1 3 Doublet Example UAV Flight Data 1 1 2 Ames Research Center Planning Meeting Late in the summer of 2003 a meeting within the Army NASA Rotorcraft Division Flight Controls Group at Ames Research Center was held to discuss the current status of CIFER and how it would be desirable to modify the program for future needs This meeting was preceded by a request to industry users of CIFER for feedback on potential changes The result of this activity was a summary of positive and negative aspects of CIFER and a tentative plan for modernizing the program and addressing some of the negative issues This section will detail some of the major points of the meeting that directly affected the course of this thesis In its current form CIFER is a collection of mathematically robust algorithms that have been tempered by 20 years of flight
93. ritefr function The example below shows the creation of a new frequency response with all 10 data arrays Again the order is fixed and as above in the getfr example not all the inputs need be provided If all inputs are not provided the user should keep in mind that CIFER will not have access to any unspecified information in the event it must make calculations with the new responses The response names do not have to conform to CIFER naming conventions though doing so will help track new responses gt gt name NEW FREQUENCY RESPONSE gt gt writefr name frq mag pha coh gxx gyy gxy rel img err 81 caselist Lists of cases present for any of the CIFER programs can be obtained using the caselist function The three calls below return a cell array list of FRESPID MISOSA and COMPOSITE cases that are present in the database The input code follows the same numbering convention found in the delete utility of CIFER Utility 13 This information can also be found by making a help call for the function and is printed in Appendix G gt gt names _f gt gt names_m gt gt names_c caselist 1 caselist 2 caselist 3 Returns FRESPID cases Returns MISOSA cases Returns COMPOSITE cases oP o o 82 Appendix C Online Help for Main Programs This Appendix contains the information that is displayed when the help functionname command is used in MATLAB FUNCTION frespid DESCRIPTION
94. rithmetic allows users to check data 16 consistency by reconstructing responses for various signals based on kinematic laws of motion Examples of this type of analysis can be found in Chapter 4 The arithmetic function uses the screen interface from the main programs but only has two screens Thus the conversion to command line interface was reasonably straightforward 2 5 Parametric Modeling NAVFIT DERIVID VERIFY The three remaining main programs of CIFER will be covered in less detail as this project does not involve them It is important to note their use as this constitutes a major portion of CIFER s potential and the direction that future work on modernizing the interface will take NAVFIT allows users to fit a transfer function to a single frequency response The interface uses a cost function to give the user an indication of how closely a particular fit matches the response data The user specifies the order of the transfer function desired and then the program iterates to find an optimum fit The interface used by NAVFIT is similar to the RMS and bandwidth utilities except it has many more input options and more loops at certain segments Additionally it was built up on code that was originally created by McDonnell Douglas Aircraft outside of NASA Thus it would be particularly challenging to adapt to a command line format This was one major factor for not including NAVFIT within the scope of this project DE
95. ross weight of 328 pounds and carries 60 pounds of payload It cruises between 65 and 85 knots up to a 15 000 foot ceiling with endurance better than five hours Figure 4 13 shows Shadow at launch Figure 4 13 Shadow 200 TUAV 42 Two sets of data were used one was a simulation of the UAV and the second from a series of flight tests on the vehicle The goals of the analysis were to investigate data consistency using the frequency response arithmetic utility and to explore the RMS and cross over characteristics of the vehicle The results of the analysis were compared to scaled bandwidth criteria for manned aircraft This analysis was the final most involved validation of the CIFER MATLAB interface 4 4 1 Data Consistency checks Checking data for kinematic consistency assures that measured data obeys kinematic laws and does not contain hidden scale factors or delays The frequency arithmetic feature of CIFER allows reconstruction of parameters not measured during flight tests from the responses of those that were measured Additionally it can be used to show whether or not data is consistent with kinematic laws There are several relations among commonly measured rates and attitudes for aircraft as shown in Equations 4 4 through 4 8 Herein it was assumed that V and Wo were small p 4 4 q 0 4 5 u a gO0 W qt Vor 4 6 v U a Uyr Wyp t go 4 7 w U a a U q Vyp 4 8 The simulation data include
96. rpart while offering new features that make modern GUIs versatile such as browsing capability and easy navigation It was determined that the work encompassed by this thesis would be a prototype or test bed for these capabilities that will result in modernized functionality for CIFER 1 2 Project Scope The goal of this project was primarily to develop a command line interface and a modernized GUI layout for CIFER It was decided to use The Mathworks MATLAB as a medium for this development as many users of CIFER also use MATLAB and a more developed communication between the two programs would be very useful The ultimate goal of CIFER modernization would be to make it independent of other programs so users do not have to purchase additional and potentially unnecessary software licenses to benefit from the upgrades Some CIFER users did express concern with the development of the MATLAB functionality because they were not also MATLAB users This concern can be addressed by a compiler developed by The Mathworks that allows MATLAB files to be compiled and run independently of MATLAB itself The MATLAB compiler that is available allows MATLAB M files to be compiled into C code which removes the necessity to have a MATLAB license to run the M files The capability for scripting and the modern GUI on which this thesis is based will be a significant benefit to CIFER users when it is fully developed It was deemed acceptable to use MATLAB
97. rst method was to create a case in MATLAB using the interface functions and then create a second identical case except in name the traditional way from within CIFER From here the frequency response results could be extracted and compared This was how the error was initially discovered Comparisons of the errors in magnitude and phase were plotted as shown in Figure 3 2 which revealed that slightly larger errors occurred at high or low frequency regions that tend to exhibit low coherence The trend was not completely true in all cases but most plots tended to exhibit that behavior Several different cases including the XV 15 data Shadow 200 data Figure 3 2 UH 60 simulation data and an example mass spring damper system were examined in this manner with the results generally being the same Figure 3 2 is a plot of the percent error as such it can be seen that these errors are small in magnitude 24 x 107 ELE_Q x 10 ELE_Az Mag dB o mn Mag dB Phase deg Phase deg N x10 ELE_PHI x10 ELE_VDOT Mag dB Mag dB Phase deg Phase deg N 0 1 1 10 100 0 1 1 10 100 Frequency rad s Frequency rad s Figure 3 2 Percent Error Comparisons In order to determine if the error was originating in MATLAB another set of tests was conducted First a case was set up in CIFER and run using different case names in CIFER and in MATLAB Then an identical case was constructed in MATLAB and again run in CIFER
98. s 48 Mag dB az Uo q Uo alphadot 90 5 a S 45 g 8 amp 4 180 Q 0 8 5 S 0 6 Fe i 8 0 2 0 1 1 10 100 Frequency rad s Frequency rad s Figure 4 20 Vertical Velocity Perturbations Figure 4 21 Comparisons with Exact 1 s Flight Data Value 4 4 2 RMS and crossover comparisons CIFER has utilities which calculate the root mean squared RMS value and some handling qualities metrics for a given frequency response The results of these calculations can be compared to handling qualities specifications such as those laid out in MIL STD 1797A or the Neal Smith Criteria The values between both simulation and flight data can be compared to check consistency of the simulation Additionally RMS calculations can be made on the time history data and compared to the equivalent frequency domain RMS values The first check was a comparison between flight and simulation full range RMS values for control inputs and output responses from the main on axis channels These results are tabulated in Table 4 4 It should be noted that RMS values are based on the input or output autospectrum for a given response and are a measure of the energy or excitation of the system 49 Table 4 4 Full Range RMS Values Flight Sim AIL 6 05 9 07 P 0 350 0 330 ELE 2 66 3 98 Q 0 193 0 281 RUD 2 36 2 94 R 0 064 0 075 WHL 1 10 1 77 R 0 064 0 075 Beta 0 047 none The RM
99. s Both methods of calculation overlay nearly precisely which makes sense as both plots are based on the same data one created straight from the time domain and one built through frequency response arithmetic Even in regions of low coherence both calculations include the same errors and thus arrive at the same answers The coherence plots vary due to a convention in 45 100 Mag dB deg Phase Coherence k Q ao Phase deg 8 ie DQ o 360 oocoecse oN A OQ the arithmetic utility that applies the coherence from one or the other of the constituent responses to the resulting response udot ax g theta verification pitch axis vdot ay g phi Uo r verification yaw axis udot_freq aes udot_time vdot_freq vdot_time Mag dB S 2 amp 1 10 100 0 4 1 10 100 Frequency rad s Frequency rad s Figure 4 16 Longitudinal Velocity Perturbations Figure 4 17 Lateral Velocity Perturbations wdot az Uo q verification pitch axis Mag dB deg Phase Coherence 0 1 1 10 100 Frequency rad s Figure 4 18 Vertical Velocity Perturbations Performing consistency checks on the simulation was useful to ascertain that the models were correctly set up and identify any discrepancies that may have resulted from hardware in the loop 46 The same checks were run on the flight data as well Measurement instruments in flight
100. s measurements of phi and theta as well as the rates and accelerations in all axes however alpha and beta channels were not provided Velocity perturbations were reconstructed from the time domain data in the FRESPID program using Equations 4 6 through 4 8 The frequency responses could then be calculated for the velocity perturbations Using 43 frequency response arithmetic data consistency can be examined for all five of the above relations The arithmetic feature in CIFER allows a single basic operation addition subtraction multiplication and division to be performed on two frequency responses Each individual response can be modified by a scale factor and a power of s if desired In the frequency domain Equations 4 4 and 4 5 can be rearranged into Equations 4 9 and 4 10 respectively thus allowing the comparison of the measured responses to the exact value of 1 s in the frequency domain Frequency response arithmetic was used to calculate the p and 0 q responses from the on axis responses for rates and attitudes Figure 4 14 and Figure 4 15 show the arithmetic results for Equations 4 9 and 4 10 and compare these responses to the theoretical value of I s Cit 5d 4 9 p 69 4 anh Se 4 10 q Both plots compare very well which is to be expected as the measured values from the simulation should be kinematically correct to start with At low frequency there is a more significant difference primarily due to the lo
101. s three choices to exit save and exit or save and run the batch job If a batch job is run the output information is displayed to screen shown in Figure A5 but users can press a key to continue using CIFER which will remove the information from the screen Once the batch job is completed additional information is displayed which can disrupt work if one has moved on while the batch job was running The final screen in the new interface Figure A10 has the three options from the original code as well as several new features There is now a dedicated window to display the output from the batch job ensuring that the information will not be lost Additionally an option to view the output log file has been added The log file contains a summary of the information from processing the case and is very useful for debugging a case that generated errors 3 2 3 Problems Encountered and Solutions The most challenging aspect of the GUI development was creating a layout that the largest number of engineers were comfortable with It was ultimately made a requirement for the new layout to mimic the old interface as closely as possible Before this requirement was set several layouts were considered that would have been a significant change to the old look and feel The driving factor was to keep the interface similar so longtime users would not have to make major 31 adjustments to their understanding of the program while adding the modern GUI fea
102. se Fewer outputs can be specified if desired Finally the printed output can be turned off by the toscrn field in the structure One other feature of the cifrms function is the ability to calculate a fraction of a full range rms value The minfreq and maxfreq fields must be set equal to use this option The printed output is slightly different and the variable outputs change slightly msv stores the full range rms value 75 rms stores the rms value for the partial range and freq stores the frequency where that partial value occurs An example call is shown below gt gt Imsv rms pts min max freq cifrms 0 name XVLATSWP FRE A0000 AIL P minfreq 0 707 maxfreq 0 707 Frequency Response Information First Freq 0 13963 Last Freq 31 41590 Number of values in frequency response 923 xxxk Mean square value results Full range root mean square value is 3 040409 Located frequency where RMS value is 0 707000 3 040409 2 149569 At frequency 3 837147 rad sec the RMS is 2 158710 cifhq The cifhq function is comprises all of the handling qualities calculations from CIFER s utility 8 as well as the utility s quick plotting linearized plotting and least squares fits There are several output variables available to store the information resulting from the function The most basic call to cifhg is shown below where the function is called to set up a template stru
103. seeeaeecaaecaaecsaeceaeenseenes 55 Figure 4 31 Scaled Pitch Category C Flight Criteria eee ecessecsecseceeceeceseceeeceseeeeeeesneeees 56 Figure 4 32 On Axis Pitch Attitude Response eeescesssecsseceseceseceseseeeeeseeeseecaaecaecsaeceaeesaeenes 56 Figure 4 33 Scaled Pitch Category A Flight Criteria 0 cee eeceeeeeeeseeeneeeneecneecnsecsaecnseenseeees 57 Figure Al Original Interface Screen Trecmecenenninn eiie e e 63 Figure A2 Original Interface Screen 2 ieni nia ee aeoaea e E EEA 63 Figure A3 Original Interface Screen Joorrncccsepoccriinaeei e E e ia 64 Figure A4 Original Interface Screen Merire eee eeeeesseeeneeensecsaeceseceseceseseeeesseeeaeecaaecsaecaecsaeesaeenes 64 Figure AS Original Interface Final Screen ui eee ccc cee cesses cee cseeeesescsseseesescssesseseesseaeaeenes 64 Figure A6 MATLAB GUE Screen lisnitcchii lian eile ie li E ETENEE EEE EE IEE 65 Figure A7 MATLAB GUE Screen 2 icicccccckccsgstden feeiiaeds i a ea RE a R RE 65 Figure AS MATLAB GUI Screen Indmu nn iiaeie e eie na ei e E a e 66 Figure A9 MATEAB GUI Soreen Aeren tnne ne a a hes a i i ele 66 Figure A10 MATLAB GUI Final Screen eee ceseceseceseceseeeseesseeseneeeseeeaeeesaecaaecaeenaeenseenes 67 Figure A11 MATLAB GUI Two Data Loading Screens eee eceeneeeneeeeeeeecnsecnaecnseenaeenes 67 List of Tables Table 4 1 Roll Gain Phase Margin Results 0 eee csesssecssecsseceseceseceseceseeeseeeaeesaaecaaecaecsaeenseenes 38 Table 4 2 Roll Bandwi
104. sely There appears to be a small offset in phase between the two after the 180 degree phase shift The offset could be due to the low coherence or possibly due to a hysteresis effect 47 Uo betadot ay g phi Uo r NT Mag dB deg g Phase Q Q o Coherence 0 1 1 10 100 Frequency rad s Figure 4 19 Lateral Velocity Perturbations Flight Data The second and final consistency check on the flight data was for w calculations for pitching motion Direct comparison of the two methods of calculation can be seen in Figure 4 20 Both calculated responses appear to have the same trend For a more exact comparison of the measured data to the theoretical solution Equation 4 8 can be rearranged to relate alpha a and q to 1 s as shown in Equation 4 12 U a 1 4 12 a U q s8 A series of frequency response arithmetic operations were used to generate the left side of Equation 4 12 which was then compared to the plot of 1 s as shown in Figure 4 21 Overall the data matches the 1 s value quite well Discrepancies in the magnitude plots can be directly correlated to spikes in the coherence The most noticeable difference occurs in the phase plot there is a small phase offset from 90 The offset is fairly constant with frequency which suggests a hysteresis effect The result is typical of airboom measurements such as were used in the flight test which tend to exhibit some amount of hysteresi
105. spective utilities Input can be specified either using a pre initialized structure or by passing in a flag which generates a template structure and a list of name value pairs to fill in the fields of the template The sections describing the individual functions have examples of both styles Each function can be called with no inputs or outputs causing a list of the information that each field requires along with its default setting to be displayed This information is provided in Appendix F Finally if the functions are called with a single output and no inputs a template structure will be returned initialized to the default settings Examples of all these commands follow in the next sections and a detailed overview of the online help information for each function can be found in Appendix E cifrms The CIFER RMS utility is a straightforward calculation for the mean square and root mean squared values of a frequency response The MATLAB cifrms function is called using a structure 74 as input and up to five outputs to collect the desired information The least complex call and the corresponding output are shown below gt gt in cifrms gt gt in name XVLATSWP FRE A0000 AIL P gt gt cifrms in Frequency Response Information First Freq 0 13963 Last Freq 31 41590 Number of values in frequency response 923 xxk Mean square value results Mean Square Value 9 244088 Root Mean Squared Value 3
106. sponse Identification is the first step in any CIFER analysis This program takes time history data and generates frequency responses based on the input and output channel measurements A very important aspect of CIFER occurs in FRESPID which is the windowing of the data The data comprises a number of discrete frequency points over a time duration FRESPID uses the CZT to average the data points of a smaller time window of data It breaks the full time history into segments based on this window and performs computations based on averages for each given window size There is a distinct tradeoff in the selection of window lengths relative to the total time history length Windows that are a smaller fraction of the total length provide a large number of averages that can more easily identify high frequency responses by countering noise effects Unfortunately low frequency identification degrades because low frequency responses tend to occur in larger intervals than the small windows encompass When the window size is enlarged low frequency identification becomes more accurate at a cost to the high frequency end due to a decrease in the number of averages CIFER solves this trade off issue by allowing the user to specific up to five different windows in FRESPID A frequency response is generated based on each window This ensures that both low 12 and high frequency content is captured from the data These preliminary responses are l
107. sponses F for ASCII files D for database Write results to the database Y or N Generate plots Y or N Control names Output names Code for windows to combine indicates a window to include indicates not to include a window Code indicating whether to clear set frealc matrix Table indicating which responses to compute requests a new response computation indicates not to calculate a response Integer array indicating which plots to generate Draw a grid on the plots Y or N Large plot Y or N Selected plot device Q MS C omprs V er S creen T alaris P ostScript Minimum frequency of each window Maximum frequency of each window composite save and run a COMPOSITE case mostly blank structure out INPUTS id cmd OUTPUTS out composite id cmd options either composite structure or string with case name Information on the details of the structure can be found with an empty call gt gt composite command to have function perform a process 1 open COMPOSITE case 2 save a COMPOSITE case 3 save case and run batch job options name value pairs to set individual data fields optional e g Casename XVLATSWP return structure for composite data structure A template structure can be returned using an empty call gt gt out composite EXAMPLE CALLS out composite returns a mostly blank template composite s
108. stored whether in a file or in the database names for the controls and outputs and the desired combinations of responses to calculate There is no conditioning or windowing of data involved which results in less error checking 2 3 Combining Windows COMPOSITE Normally the optimization of window sizes in the FFT calculation would be a very time consuming process For a four input nine output system there would be thirty six responses that would each have to be individually optimized CIFER employs COMPOSITE for composite windowing to combine the individual windowed responses from FRESPID into a single combined response thus automating the optimization It uses a nonlinear least squares optimization of a cost function based on the auto and cross spectra to combine the window data 14 Thus the low frequency content of larger windows is combined with the high frequency content of smaller windows into a single response COMPOSITE is set up almost identically to MISOSA in terms of the screen interface There are some very slight differences in specifying the sources of data but otherwise the interfaces for the two programs gather the same information Since the required input information for each of the three main programs is of similar nature it was easy to standardize the general layout of the command line interface created to drive them in MATLAB in terms of variable names structure and error checking 2 4 Analysis Utiliti
109. sts New Jersey Prentice Hall 1997 Thurling A J Improving UAV Handling Qualities Using Time Delay Compensation M S Thesis Air Force Institute of Technology 2000 Tischler M B Cauffman M G Frequency Response Method for Rotorcraft System Identification Flight Applications to BO 105 Coupled Rotor Fuselage Dynamics Journal of the American Helicopter Society Vol 37 No 3 pgs 3 17 July 1992 62 Appendix A Screen Layout Comparison This Appendix contains a series of screen shots that show each screen in COMPOSITE for both the original CIFER screen interface and the new GUI interface A more detailed description of the differences can be found in Chapter 3 Old Interface FY Fm COMPOSITE 1 KOK KKK KKK IK KKK KKK KKK KKK KKK KK KKKKKKEKKKKKKKEEK Program COMPOSITE Multi window Averaging kakkkkkKkkkKkKkKkKkKkKkKKKKKKkKKKKKKKKKkKKKkKKKKKKKKK For COMPOSITE initialization enter a CASE name Press the PF3 key to turn off prompting Press the PF1 key to accept a screen PF2 for options CASE XVLATSWP AIRCRAFT 703 Figure A1 Original Interface Screen 1 a E T Case COMPOSITE 2 Id Window composite for Tiltrotor Input prefix XVLATSWP Output prefix XVLATSWP Source program name MIS MIS or FRE Read frequency responses from D F ile or Df atabase D b output Y Y N File output N N Plots Y N Number o
110. the future The GUI was written to work with the command line interface using graphics to gather the information to be sent to the command line functions without the user needing to know how the command line functions work It is important to note that the GUI programming does not stand alone as the command line does and is to demonstrate how advances in GUI programming can make CIFER more user friendly 28 3 2 1 Development Process The most difficult aspect of developing the GUI was adjusting the look and feel to appeal to the widest audience of potential users The iterative method used for the command line development was employed for the GUI as well with NASA engineers and programmers included during the design process Several initial concepts were sketched on paper before any code was written When ideas were put into code the first programs were designed only around the layout of the GUI lacking any real functionality This allowed reviewers a clear idea of what the finished product might look like Once the design for the layout was sufficiently refined the functionality was written into the code As the functionality was based on the command line interface the focus of the GUI development was the layout The GUI was largely built up around the command line code that already could interface with CIFER Data entry in the GUI was largely handled by toggles and text fields The error checking from the command line was supersed
111. tion of the error signal to the input and feedback signals for a conventional feedback setup as depicted in Figure 4 7 Figure 4 8 shows the plot of the error response to stick input compared to the response solved with the MATLAB version of CIFER utility 9 using Equation 4 3 The results overlay precisely which was expected given that the data comes from a simulation 37 Command LATS fan S Figure 4 7 Feedback Block Diagram Mag dB deg S a Phase Coherence Figure 4 8 Error Channel Verification 4 2 4 3 Calculated Error LATS Measured Error LATS 10 Frequency rad s 100 Using the MATLAB version of CIFER s utility 8 for crossover characteristics the feedback response to error in the lateral axis was analyzed A time delay of 0 0402 was incorporated into the CIFER crossover analysis because the time delay was not present in the simulated flight recordings This modification is not entirely correct because it puts the time delay over the entire system as opposed to just within the forward loop where it actually occurs It is however a reasonable approximation Table 4 1 shows the results from CIFER as compared in Unix and from Matlab to those printed in the training course manual There was no discrepancy between the two difference CIFER calculations Table 4 1 Roll Gain Phase Margin Results CIFER From Unix CIFER From MATLAB COND
112. tions of the data that would not be used in analysis due to low coherence values The analysis provided a good illustration of the capability introduced with the ability to script the functionality of CIFER The scripts that ran most of the analysis and generated the plots would have taken far longer to enter by hand into the CIFER interface Many bugs and errors with the MATLAB code that might not have been found till much later or at all were uncovered by the extensive use of the functions during the analysis In addition to validating to the programming efforts the analysis tasks provided valuable experience to the author in real world application of controls and handling analysis The UH 60 example provided important insight into open and closed loop handling qualities The work on Shadow illustrated the need to verify that measurement devices are correctly installed and provided good experience in determining the accuracy of a simulation as compared to flight data 59 5 3 Future Work The functions created for this thesis are a significant step forward for CIFER but there is significant work that must still be completed in order to ready the command line and modern GUI capability for commercial release which is the ultimate intended goal Command line primitives for the remaining programs and utilities need to be created as well as a full graphical interface While DERIVID and VERIFY are too interactive and complex to be encom
113. tructure out composite TEST 1 opens case named TEST returns composite structure 86 out composite TEST 2 casename TEST2 winon BAT oN I Ie TE Ay saves TEST as TEST2 and the window selection is changed the structure is returned composite TEST2 3 sends case TEST2 to batch nothing is returned NOTES A totally new case must be specified via a structure not by the desired new name i e composite NEWNAME 2 will NOT save a blank case entitled NEWNAME into the database DESCRIPTION OF FIELDS IN INPUT STRUCTURE SCREEN 2 id casename Composite Case name id caseid Descriptive text for this case id casein Frequency response prefix input id caseout Case part of freq resp name for output id inpgm Input program FRE or MIS id source Source for input frequency responses F for ASCII files D for database id savdb Write results to the database Y or N id plot Generate plots Y or N id outpts Number of points in each resultant freq resp id controls Control names id outputs Output names id winon Code for windows to combine indicates a window to include indicates not to include a window id winlen Window length in seconds SCREEN 3 id frall Code indicating whether to clear set ICOMP matrix id frcalc Table indicating which responses to compute requests a new response computation indicates not to calculate a response SCREEN 4 id plotopt Integer array indicating
114. tures that might shorten the learning curve for new users The second major challenge lay in generalizing the code structure to allow other programmers to easily adapt the code to work for other CIFER programs which was accomplished using the aforementioned added navigation functionality These navigation tools were set up to work for a general series of windows For this project these were for COMPOSITE but if windows for another program were created they could be easily linked This concept was illustrated by the lead programmer for CIFER at NASA who was able to adapt the code to the MISOSA program in a few days as opposed to the initial development which spanned several weeks The development of the GUI was a significantly smaller undertaking than the development of the command line The GUI largely added to and enhanced the already present functionality of the command line Thus there were fewer technical problems associated with development The practices set in place from the work on the command line interface continued to be employed for the GUI development 32 Chapter 4 Validation and Application In order to confirm that the new code would accurately move information between CIFER and MATLAB many tests were conducted In the early stages of the code development various sample cases provided with CIFER installations such as the XVLATSWP case mentioned previously were used Also at this stage a simple mass spring da
115. w coherence The difference suggests that the values from the CIFER response are unreliable due to little or no input at those frequencies This is true of the higher frequencies where coherence also degrades Another cause of these discrepancies could be a result of the hardware in the loop which could introduce sensor and actuator noise and biases oO g g 90 F o 45 QD amp 180 Coherence Roll Verification phi p 1 s Pitch Verification theta q 1 s gt 1 s calc 1 s_exact Mag dB 0 8 0 6 0 4 0 2 0 1 deg Phase Oo 0 8 S 0 6 S 04 8 02 1 10 100 0 1 1 10 Frequency rad s Frequency rad s Figure 4 14 Roll Angle to Rate Comparison Figure 4 15 Pitch Angle to Rate Comparison The next check using the simulation data was performed for Equations 4 6 through 4 8 Frequency response arithmetic was used to combine the on axis responses of rates and attitudes to calculate the velocity perturbations in the frequency domain from the appropriate frequency responses These were compared to the corresponding velocity perturbation responses that were initially reconstructed in the FRESPID program using the same math on the time history data Equation 4 11 shows an example of how the frequency responses were combined using arithmetic a lq 4 11 r S ate These comparisons are shown in Figure 4 16 through Figure 4 18 using a forward velocity of 65 knot
116. which plots to generate id grid Heavy grid on the plots Y or N id lrgplot Large plot Y or N id plotdev Selected plot device Q MS C omprs V er S creen T alaris P ostScript 87 Appendix D Structure Field Specifics for Main Programs This Appendix contains the information displayed through an empty call to any of the functions Default settings are shown within a curly brace gt gt frespid Structure fields casename caseid controls outputs caseout crosscor savdb plot evntnum flghtnum strttim stoptim source thdt biasflag thfile conchnl1 conunit conscfac outchnl outunit outscfac frall frealc conditioning condunit savconth outpts winon winid winlen wininpt winoutpt winde minff maxff plotop plotde gri lrgplo plotde are string 8 characters ae l string 60 characters cell array 10 of a 3 in cell array 20 of arent string 8 characters ry N ry N J ry N scalar array 10 0 scalar array 10 0 positive scalar array 10 positive scalar array 10 integer value 1 positive scalar 0 ry N OO a cell array 10 of strings cell cell pos cel cel pos y cell scalar array cel yv pos cel cel pos positive i i i i array 10 5 of strings array 10 of strings tive scalar array 10
117. would be required for verification Flight Simulation ao 2 D i Q g N 270 a A 1 10 Bandwidth rad sec Frequency rad s Figure 4 31 Scaled Pitch Category C Flight Criteria Figure 4 32 On Axis Pitch Attitude Response A third bandwidth criterion for Category A flight encompasses non terminal phases that require rapid maneuvering precision tracking or precise flight path control Examples include air to air combat weapon delivery launch reconnaissance or terrain following While Shadow 200 is not a combat aircraft it might be expected to perform aggressive maneuvers during its transit to and from observation targets Figure 4 33 shows Shadow to be well into the level 2 region for Category A flight The level 1 region scales up to a very high bandwidth which may be unreasonable Given that the spec is derived from fighter aircraft data a scaled down fighter style 56 100 UAV might conceivably achieve bandwidths necessary to yield level 1 handling qualities by these scaled criteria 0 16 0 14 0 12 o b bag o a Phase delay sec o O S F S So X o 0 5 10 15 20 Bandwidth rad sec Figure 4 33 Scaled Pitch Category A Flight Criteria 4 5 Summary of Validations The extensive series of validation examples suggest that the CIFER MATLAB interface from the command line works very well No major fundamental errors are still present in the cod
118. y many of the features that make CIFER accurate and robust contribute to making it difficult or tedious to use at times There is a steep learning curve to become familiar with the myriad of functionality the program offers While the program is robust it is not always instructive in alerting users to the nature of a problem error messages can often scroll too quickly and are cleared from the screen before they can be read The status of running batch jobs can be difficult to assess Additionally moving data from responses and plots into modern programs such as MATLAB or Igor can be very challenging One almost universal concern with the program is the interface Modern engineers are growing less familiar with the Unix based Curses interface and much of the user feedback requested an update to the interface Even to experienced users the same linear interface that gives CIFER its power and robustness can be a serious hamper data fields must be retyped for new cases with limited cut and paste functionality Another major concern is that there is no way to script the processes of CIFER setting up a series of multiple cases for an involved flight test can take days of repetitive data entry One conclusion of this meeting was that a modern graphical user interface GUD and a way to use command line calls to script CIFER processes should be developed A thoughtfully laid out GUI should be able to retain all the robustness of its Curses counte
119. y response cell array containing names of cases for specified program 99 Appendix H Mass Spring Damper Case Example This example is for a simple second order mass spring damper system modeled in Simulink as a transfer function A frequency sweep simulating noise was run through the model and the input and output data recorded into a time history file One case was set up entirely in CIFER and then a second was set up and run from MATLAB The commands shown below were used to run the case from MATLAB Calls to the CIFER MATLAB interface functions have been highlighted in red This example is provided to give a very simple scenario to set up cases using the MATLAB interface CIFER does not come with this example so the simulation would have to be created and run in order to generate the data necessary Assign a blank frespid structure EIn frespid thename MASSSPRG Fill in all the necessary information to make the case in casename thename f in comments mass spring system in caseout thename f in crosscor Sys Time history selection parameters _in source 5 _in evntnum 1 aly _in flghtnum 1 1 an thdt 0 01 in thfile 1 massspring CIFERTEXT channel definition parameters _in controls 1 IN _in outputs 1 OUT _in conchnl 1 1 IN _in outchnl 1 1 OUT Frequency response selection parameters f in frealc 1 1 a Hh Fh Hh Hh o Hh Fh

CIFER - MATLAB Interfaces: Development and Application

Contents

Download Pdf Manuals

Related Search

Related Contents