SALSA - User's Manual
Contents
1. Associated State NORMAL OP Ooo Nita PN State Specify Class Record T Nostate T Notte ME Class Name Class Name CL1 Class Number 1 Class Record nje oo l FIGURE 30 ASSIGNMENT OF CLASS TO STATE 15 The table may be viewed or edited with the View Edit Mapping button The user can also view the assignment of states to the individuals by clicking the States button Figure 31 FIGURE 31 ViEew EDIT MAPPING amp STATES BUTTONS 6 LIST OF STATES 1 NORMAL OP 5 RECOVERY HIGH WWATER ALARM BLOCAGE SHUT DOWN aE n Ja OI Q NI STATES uw I Ki N I LEGEND E Normal H Transition DS r e s locala aata i j Mean Bi eer I I I I I I 12 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 enn _ B Faut gt BB No State CURRENT STATES ndividuals View Edit Mapping _ FIGURE 32 CURRENT STATES GRAPH SALSA allows saving the different contexts used active descriptors presence function for quantitative descriptors connectives and exigency level as well as the classification results for future use The context will be saved in a binary format with a CON extension Whereas classification parameters classes and states will be saved within a binary format with a CLA extension See Figure 19 for location of save button2. Export CLOSE FIGURE 26 ANALYSE WINDOW MEMBERSHIP GRAPH l l I l l l l l l l l l l l l l I l l a a a e N l 10 12 14 16 16 20 22 24 26 20 30 32 34 36 36 40 42 44 46 4 50 52 54 56 58 Individuals CLASS LEGEND FIGURE 27 MEMBERSHIP GRAPH 3 1 7 Mapping Classes to States When a suitable classification is obtained the user must assign the different classes into representative States This is done using the Mapping to States button located at the right bottom side of the Current Classification graph The user must first of all create the list of possible significant states by clicking at the Create List of States button see Figure 28 14 Mapping Classes to States CREATE List of States FIGURE 28 MAPPING CLASSES TO STATES WINDOW The user can add and remove the number of representative states of the process Once the list is completed the user can close the window by clicking the OK button see Figure 29 State List Oi HORAL Oj HIGH WATER D3 ALARM 114 BLOCAGE Oy5 RECOVERY Add Remove DK FIGURE 29 CREATING THE LIST OF STATES After the list of states is created the table of classes and states must be filled up A class can be mapped into a state from the list by double clicking into the cell of the corresponding class Figure 30 aat EDIT List of States Double Click on a cell to edit a Class Record SsMapping Classes to States
3. Situation Assessment using LAMDA claSsification Algorithm User s Manual 2009 v 1 0 CONTENTS 1 ABOUT THIS USER MANUAL cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccecs 1 2 SOFTWARE INSTALLATION crien n a a a E a 1 Z V INTRODUCTION lt c 2260202226004226 E E EEA 1 252 TECHNICAL DATA 2essasecresisdireredenesra EEO OEE veces EEEa a a AMECA NAAN EREM AREN 2 ZS SALSA SETUP o 040 406 24 cece ANR NCA EA ici lows es tendered ese AEEA 2 3 GETTING STARTED SALSA MAIN SCREEN ccccccccccccccccccccccccccccccccccccccccccccccccccecs 2 3424 5ALSA OFFLINE 2 cs riticetnaririescsre aa SCRC E ARANAN A a a aaa 3 3 1 1 LOADING THE TRAINING DATA FILE IN THE SPECIFIED FORMAT ccccccccccccccccccccccccccccccccccccccccccccccs 3 Sl le ROR Mae opece LI Ol chetecacstoscoteatatescasetenacedeeesn APAIA ACETA METALL aS 3 3112 LOdE a TAINS Er Aa N t ETO 0 TEA Tt GRELA OLAR LAA 60 CANADA AUT LTD 076 4 3 1 2 CHOOSING THE CLASSIFICATION ALGORITHM QUANTITATIVE DESCRIPTORS cccccccccccccccccccccccccs 5 3 1 3 LOGIC OPERATORS AND THE EXIGENCY LEVEL ccccccccccccccccccccccccccccccccccccccccccccccccccccccscccccsceccs 6 9 ASA EOE IE ODEN AU ONS eia E AS NORA DESSA SG ROCAS eset 6 Bod EXIPENICY LEeVel 254444 6250022442922649 EA atte erased ETE ANET 6 3 1 4 CLASSIFICATION MODE 1442422 dds eraa a recehetetd bates aviedde seine eee 6 3 14 SOL Leane 222225 RL a E cred REES ELE dE 7 3142 ActvESupervised Learning ar
4. REFERENCES 224002 6605050460506064 O teneaemeemsue sie E E 21 FIGURES TABLE Figure 1 SAULSA Main Screen OPTIONS vscwica snes sess usesccusveetcccavicaxeusenccocd vee scdutveetcetavicaxeuiensivonistnscdateaetiouseneucads 2 Figure 2 SALSA OFFLINE Main Sr eG vavkemaeshnt cous chokexeeehel does shobeomvchalcoeeeholeuaeetaees 3 Figure 3 Example of the data cdat File wececersceseevcceocevietencesnceuisuargtiecesardecauaceeweaaede CGA ECE PACA E 4 Figure 4 New Context amp New Population buttons ccccccssssssccccsssssccccessescccceessesccsseessscceseassscesensees 4 Figure 5 Loading a New Context Induce OF R triCVC ssscccssssccccnssscccssscccnssscccessseccnesscccaesscssessseceoees 5 Figure 6 Loading a NEW PO DUIATION wx cececdececsceeececes ceoeepuccawcesnteyecesncdewseeuseedaueteuveesetedeieinteein i fase munesenteoneees 5 Figure 7 View Edit File amp Imposed Classification WiNGOWS sssssssssssssssssssssssssssssssssesssssssssssssssssesees 5 Figure 3 Presence EUNCHON SEICCUION sorsoran nnn E A esa ae ee ee ome 5 Fiz re 9 Connectives amp Exigency selection xcvecsestes sac ensateseneiasstedute sul eevdacdiaSuicodvesntedeietusedscesstacsheauiegelesseees 6 Figure 10 Classification Mode SCI CtION ccccssssssccccnssssccccnsssscccceasssscccenasescccceassscesenasssccsseasssscesensees 6 Figure L1 5Se lFl arning Selector snan tuaceectteaverdc R REA ude Oe 7 Figure 12 Automatic Self Learni
5. T norm amp T conorm 3 1 3 2 Exigency Level The user may also use mixed connectives of the same family by choosing an exigency level a between o and 1 Figure 9 In recognition we call the exigency level and in self learning it is called selectivity level By changing the value of a different partitions based on the same data used may be obtained Thus as the value of a increases more classes will be created or in the case of recognition a greater adequacy is required of a measurement to thee assigned to a pre established class Connectives Probabilistic Exigency Index F 1 000000 FIGURE 9 CONNECTIVES amp EXIGENCY SELECTION 3 1 4 Classification mode When a file has been loaded for the first time the classification mode will be selected according to the characteristics of the file Tee aa ge gea Ser Sage Sage Sane Seer Sage Sace Soee Sage Sage Saar Saer Sede sev Classification Mode Supervised Learning Self Learning v Supervised Learning Recognition FIGURE 10 CLASSIFICATION MODE SELECTION 3 1 4 1 Self Learning Also known as unsupervised learning meaning that there are no pre defined classes no preliminary knowledge of how the elements are assigned SALSA creates a group of classes using as base the values of the descriptors from the individuals of the data file that has been loaded Manual Automatic Max Variability 1 000 Total terations 0 Max erations 1
6. To EXIT the online screen the user must FIRST STOP the online connection with the STOP button and then EXIT the screen with the EXIT button 4 APPENDIX 4 1 ONLINE DATA GENERATOR This application permits to generate online data from stored data files It can be used in order to validate SALSA Online stage The Online Data Generator is a standalone executable that may be installed in the following way In the InstallationSALSA04v1 Folder look for the sub folder Package Click the setup exe file 2 Installation de SALSA File Transfer Generator Commence l installation en cliquant sur le bouton ci dessous Cliquez sur ce bouton pour installer le logiciel 54154 File Transfer Generator dans le dossier de destination sp cifi Dossier d 54 5403 2bat Changer de dossier Quitter l installation FIGURE 46 ONLINE DATA GENERATOR SETUP WINDOW 20 Once the installation is completed it is possible to generate online data from a file gt SALSA FILE GENERATOR ABSXNCL_02 04 dat gt Cet Data File yd EJ 5ALSA03v2bat Current datum a Current Class FIGURE 47 ONLINE DATA GENERATOR MAIN SCREEN First of all choose the data file with extension DAT and click the Set Data File button see Figure 47 Choose the folder where the online data file must be created it must be in the DataTr folder created by SALSA SALSA04v1 DataTr and click the Set Folder button The RUN button is enabled When clickin
7. 1 Min 5 24 Current STATE of the descriptor ACTIVE 3 Name 02 Ep ct Type WON ORDERED GUALITATIVE Label 1 value G Label 2 value L Label 3 value H Curent STATE of the descriptor ACTIVE 4 Name DZ ValF Type QUANTITATIVE Max 17 66 Min 0 507 Current STATE of the descriptor ACTIVE FIGURE 16 CONTEXT DESCRIPTION WINDOW By clicking the View Population button all active descriptors are graphically displayed on the lower graph of the main screen There is a graph for quantitative descriptors where for each individual its normalized value is shown and at the right part of the graph the names of the active quantitative descriptors are displayed For qualitative descriptors modalities corresponding to each individual are shown in a graph The user may change from one graph to another using the corresponding option View Qualitative Descriptors or View Quantitative Descriptors which is located at the right bottom corner of each graph See Figure 17 amp Figure 18 For qualitative descriptors we represent each modality with a similar but different colour VETE vos 033 0 8 0 7 0 6 0 5 0 4 35 22 inSl D4 02 Valf D6 T1_ValF 0 0 DS T2 ValF 0 1 a Te DIO T3_ValF 12 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 49 50 52 54 56 58 w Active Quantitative Descriptors Individuals View Qualitative Descriptors Value imi i ee i Di Active Qua
8. FIGURE 1 SALSA MAIN SCREEN OPTIONS 3 1 SALSA OFFLINE Figure 2 shows the main screen for SALSA offline learning stage The following sections give some guidelines of the features and use of this stage 2 OFFLINE SALSA Situation Assessment using LAMDA claSsification Algorithm SYSTEMBUSY __ CONTEXT POPULATION New View CLASSIFICATION Retrieve 0 1 1 1 5 10 45 50 55 60 65 70 75 80 85 90 95 Connectives Probabilistic v SA Individuals Happing to States Current Classification j Exigency Index 1 000000 1 0 0 9 0 8 0 7 0 6 3 T 0 5 gt 0 4 Manual Automatic Max Variability 1 000 Total Iterations 0 Max Iterations 1 10 15 2 2 3 3 4 4 55 8 765 g e Active Quantitative Descriptors Individuals View Qualitative Descriptors FENT lt i O WIDTH 50 1 Gm Transitions Compare STEP FIGURE 2 SALSA OFFLI NE MAI N SCREEN 3 1 1 Loading the training data file in the specified format 3 1 1 1 Format Specification The format for the input training files the context data base and the population to be classified is a text file with a DAT extension Lines can be blank ones comments MatLab style always preceded by or observations These observations or individuals present values for the context descriptors in columns separated by blank spaces or tabs and may have an additional field if we are interested i
9. FIGURE 11 SELF LEARNING SELECTORS The Manual Automatic selector Figure 11 gives the possibility to make an automatic scanning combining the three principal parameters of LAMDA algorithm The result is a table with the number of classes generated by LAMDA using self learning and the partition s quality given by the quality index CV proposed by Isaza 2007 see Figure 12 The partition with the highest quality value for each MAD function is highlighted number of classes generated and CV index value Automatic Classes Partition Quality Automatic Classes Partition Quality For each Presence Function the classification with For each Presence Function the classification with the highest Quaility is highlighted the highest Quaility is highlighted 0 0160281 0 0168595 0 0167566 oa 0 0157978 0 0163123 0 0162594 O MINMAX 0 6 00150315 0 0153416 0 0153378 0 0625193 0 0902129 0 0591460 Sez 0 1170478 0 17215640 0 128045 EXIGENCY LAMDA1 Lampas GAUSS 1 EXIGENCY LAMDA1 LAMDA3 GAUSS1 0 2 2 1 a ae 0 0008145 0 0005536 0 0172148 o 4 0 0008362 0 0006042 0 0171556 PROBABILISTIC 0 PROBABILISTIC 0 6 0 0008794 0 0007021 0 0170087 i 0 0010069 0 0009750 0 0165120 se at 0 0197041 0 0180866 0 0153527 FIGURE 12 AUTOMATIC SELF LEARNING RESULTS CLASSES amp PARTITION QUALITY The other two selectors in Figure 11 that may be changed by the user when uns
10. Figure 24 Recognition Results WIndOW ccccececcccccecccccecececccccccccccccccccecccccccccococecococceececoceococoneoooanees 13 Figure 25 Reference Classification Graph eececceeeccccceccccccccccccccccccccccccccccccccccocococooocococeeocsooseooovenees 13 Figure 26 Analys WIndOoW 2 2 20250220 6406049365004004 E C 68 200800 C L 08009900 80 bended E anaes teens 14 Figure 27 Membership Gla Pieces iccidevecdeheactesviesciseiescte AEREE 14 Figure 28 Mapping Classes to States WiINdOW scscccssssssecccesssssccccesssscccceassescccensssccseeesscesseasessceseneees 15 Figure 29 Creatine tie list OF States nana SOTA swede cea bnsdeseasso sateen aGedeccenesuteteadodiaasneeeles 15 Figure 30 Assignment Of Class to St te rsi 0 EENE DEEE TEE E 15 Figure 31 View Edit Mapping amp States DUTtONS sssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssssees 16 Figure 32 Current States Grapini a n r ince veel tnndevea O OET O A 16 Figure 33 SALSA Online Maln SEl een vciicaiseidesos seesceosenel sass capvenscea sd EE ATA Eaa E 16 Figure 34 SALSA Online COMTIGURATION sie E EA eects RANA AS 17 Figure 35 List of Online Quantitative Descriptors seesessseeecosoeeecososeccosoesceossececossscecossseeeossoeecossoeecessoe 17 Figure 36 Ready to Send Online CONPBULAT OM minesinin eneen EEEa cles NETA AN S RS 18 Figure 37 Reddy TO RECEIVE ONINMe Cate saisctc ches tcsce ae cecli terre tic eteehiue E s
11. From the installation Folder InstallSALSAo9 Beta double click at the setup exe file and let you guide by the installation wizard 2 Once SALSA application has been installed verify that the directory SALSA09v1 has been created 3 Verify that the sub folder DataTr has been created in the SALSA09v1 directory This folder is required for the online data transfer between SALSA and another application running online 3 GETTING STARTED SALSA MAIN SCREEN When SALSA is launched Figure 1 shows its principal window There are three main options to choose OFFLINE Learning Stage This concerns the design and construction of a classification system from a set of experimental data It consists of two main stages a learning stage unsupervised or supervised and the assignment of the resulting classes into significant functional states Active participation of the process expert is required ONLINE Recognition Stage Using a previously generated classification system the current functional state of the process will be identified QUIT This option exits the toolbox To exit the application the user must first close quit the option he is working on one of the above Then exit by clicking the QUIT button in the principal MENU window SALSA Situation Assessment using LAMDA classification Algorithm x OFFLIN ONLINE CIT Situation Assessment using LAMDA claSsification Algoritwmn
12. SALSA will take into account a 0 or if a value is greater than the maximum value SALSA will replace it with a 1 and the process will continue The user may view each context descriptor separately and in a detailed way as well as the values of each individual Figure 15 It is possible for the user to change the descriptors he wants to use for learning and classification he can enable and disable a descriptor for classification Descriptor Details Descriptor Details INACTIVE R ACTIVE INACTIVE E ACTIVE Quantitative Descriptor Descriptor 8 T2_ValF Individual ee Number of individuals Normalized Value 0 483 H H Individuals Modalities FIGURE 15 QUANTITATIVE amp QUALITATIVE DESCRIPTOR WINDOWS For qualitative descriptors a histogram of the number of individuals for each modality is shown in Figure 15 as well as a small window that gives the name of the modality and the exact number of individuals For quantitative descriptors a graph representing each individual with its normalized value is shown There is also a text window with all the information about the loaded context see Figure 16 amp Context Description The present context has 10 descriptors and they are 1 Name CO2 Ep ct Type NON ORDERED QUALITATIVE Label 1 value G Label 2 value A Label 3 value F Current STATE of the descriptor ACTIVE 2 Name CO2 ValF Type QUANTITATIVE Man 2
13. This file has the name d_in dat It contains one single line in the same format as for the training files see Figure 39 Once SALSA reads the d_in dat file it destroys it and when recognition is made an output file is generated d_out dat see Figure 40 The application in charge of sending the online data must destroy this file and create the d_in dat file again to tell SALSA that a new message has arrived Data Transfer Folder E TatianaK SALSA Programart5ALSA JANCS DataTr FIGURE 38 PATH FOR DATA TRANSFER P d indat Bloc notes Fichier Edition Format Affichage gq h 3 387 low high 670 872 o p 7 446 0 006 q a 159 511 0 957 FIGURE 39 FORMAT FOR THE ONLINE DATA FILE P d_out dat Bloc notes Fichier Edition Format Affichage 02 06 2004 11 39 49 NORMAL NoRegul 1 122622 05 FIGURE 40 FORMAT FOR THE OUTPUT FILE BY SALSA 18 View Configuration Send Configuration Start Time 04 05 2006 15 59 25 FIGURE 41 STOP RECEIVING DATA The user may stop receiving messages with the Stop Data button Figure 41 3 2 3 Receiving Online Data The online screen offers the visualization of the incoming data the current state and the state of the plant at the instance of an arriving message It has a historical table with the time at which there was a change of state with its associated class and corresponding GAD Global Adequacy Degree 3 2 3 1 Visualizing incoming data The use
14. a RES extension It is possible to save the classification result as the reference one by clicking on the Save Ref CL button The user may choose the name of the classification and he can decide to set it as the reference for comparison with other classification results 3 1 6 3 Classification Analysis The ANALYSE button shows the resulting membership matrix of each individual for each existing class Figure 26 This matrix can also be graphically viewed by clicking on the View Membership Graph button Figure 27 The user may save this matrix in a CSV format by clicking on the EXPORT button The analysis window displays as well the partition quality index value which allows the user to make a comparison between other classifications The higher the value of the CV index better the partition s quality 13 MEMBERSHIP MATRIX Class 1 Class 2 1 O 4466201954 0 2556 704045 0 Z 0 4000906090609060 0 2500000000 0 3 0 5914618193 0 2619257885 0 4 0 4500000000 0 2952448965 0 5 0 4000060060060 0 25060606000 0 6 0 3500000000 0 295244838965 0 0 2549356474 0 250006060000 0 8 0 2500000000 0 2952448965 0 9 0 2500000000 0 3437500000 0 19 0 2500000000 0 4375000000 a 11 0 2500003558 0 4687500000 ae Ea 4 Class Conectivity Classes Adequacy 04687500000 gad NIC 02587755221 Cutting Alpha a Ee Partition lt 0 500000 Population Covering 0 2501315196 Quality 0 0260515295 View Graph
15. and if the loaded population has pre defined classes a Results Window is displayed with different information Figure 24 number and percentage of individuals not recognised number of individuals with pre defined classes number and percentage of individuals badly recognised and the list of the individuals badly recognised 12 Recognition Results Recognition of Pre Classified Individuals Recognition Type allows INDIVIDUALS in NIC Total number of NON Recognized Individuals 3 Individuals with Pre assigned Classes 54 Number of NON recognised individuals 1 1 85 X Number of Individuals Bad Recognised 4 gt 7 56 X List of Individuals Badly Recognized 19 31 32 4 FIGURE 24 RECOGNITION RESULTS WINDOW When a population with pre defined classes has been loaded the View Prof CL button is enabled Figure 21 By clicking this button the pre defined classes for each individual are displayed Figure 25 Reference Classification 4 u in zai w od 3 LI Zt E E E E S 88 08 0 8 8 8580 8 88 8 88 8 8 T a ee eee E E E E E E E E O l l l l l l l l l l l T T l l l l l l l l I l l l I I l q2 4 6 8 10 12 14 1b 18 20 22 24 26 28 30 32 34 Gb 38 40 42 44 46 46 50 52 54 56 58 Reference Classification Individuals FI GURE 25 REFERENCE CLASSI FI CATI ON GRAPH SALSA automatically creates a result file where each individual has its corresponding class This file is a text file and has
16. cie enamel CL A PEER 00 eeabiade mae E 7 DAS Pattern Recognition 2053022224004272 EE E EO 8 3 1 5 DATA PROCESSING TREATMENT AND VISUALIZATION ccccsccccccccccccccccccccccccccccccccccccscccceccecces 8 3 1 6 CLASSIFICATION RESULTS AND VISUALIZATION ccccccccccccccccccccccccccccccccccccccccccccccccccccccesccccecs 10 3 1 6 1 Resulting classes characteris tiCS naina a a N a S 11 3 1 6 2 Recognition ResUultS cccccccciceciccccccccccccciceciceccccccrcocircocccoscocococosocooooenooenonenenenns 12 3 1 6 3 Classification ANalySiS ccccccccssscccccessecccseseccceeecccceeesecessuececseneceeeeeeeceesanecessunecesseggeceeeeees 13 3 1 7 MAPPING CLASSES TO STATES c2u3c254626440020 crt 4404640000003103 TANIE DENEAN r AANDRA 14 3 2 SALSASONLINE wercita aa aaraa 16 3 2 ONLINE CONFIGURATION vorisini nianna a A OAE AENA 17 3 2 2 ONLINE CONNECTION sssesscssssscccccssosscccccsoseseccosseseccccosssesecccssssssccsossccecocsssssseccssseseseesos 17 3 2 3 RECEIVING ONLINE DATA eesessssssccsssssscccccsosescccossececocsossescccccssssscccscssececocsososseccssseseseesso 19 3 2 Sls VISUANIZING NCOmMNE d taiensis 009 PAORS ASARD ARLE BAA CARIE A SARCA 19 3 2 3 2 VISUANIZING TheTCTurrent Stat 2 2 5222022952092 84605004 a ROA Al bE TARA UET cn eN 19 a APPENDIX arenar waves E co t d ANO A 7 S 2D A an OV DA D XA J As nn 20 4 1 ONLINE DATA GENERATOR cccccccccccccocoooooooococcccccocececcsesseseseseceeueueueueesess 20 5
17. dat POPULATION New ATA_TEST Gazeifieu_Episodes Apprent_5V2D dat CLASSIFICATION Retrieve Save _ lt hot saved sues eee Presence Function E Gauss 12 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 CEREREA Probabilstic v SEE e Individuals Mapping to States Current Classification pping Exigency Index 1 000000 tas 10 Classification Mode Self Learning vi 0 3 7 lt N Re Class Profile Save Ref Cl 0 8 Class Details view Ret ci 0 7 0 6 7 Manual M Automatic Max Variability 0 300 0 5 Total terations 2 Max iterations 4 0 4 i 0 3 TARN f gt zie df Active Quantitative Descriptors f Yt H 0 2 be D T1_ValF T1 Ya 0 0 D8 T2_ValF 0 1 DEFA 1 1 1 I 1 1 1 1 D10 T3_ValF 12 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 Active Quantitative Descriptors Individuals View Qualitative Descriptors E ees poe op z ig AA 2212 ai PRINT DON WIDTH 50 4 Transitions Compare N init Indiv 1 000 FIGURE 20 CLASSIFICATION RESULTS USING SUPERVISED LEARNING 3 1 6 1 Resulting classes characteristics Once a classification has been made SALSA graphically displays the class assigned to each individual Figure 20 and by clicking on the Class Profile button Figure 21 the profile of each class wi
18. g this button the program will create the d in dat file the information of this file will be displayed in the current datum window Online Data is generated every two seconds When SALSA creates the output file d_out dat the program will display its contents in the Current Class window and will destroy this file in order to generate a new d _in dat file with a new datum To quit the application click on the EXIT button 5 REFERENCES Aguilar Martin J and L pez de Mantaras R 1982 The Process of Classification and Learning the Meaning of Linguistic Concepts In Approximative Reasoning in Decision Analysis p 165 175 Isaza C 2007 Diagnostic par techniques d apprentissage floues conception d une m thode de validation et d optimisation des partitions Ph D dissertation INSA Kempowsky T Aguilar Martin J Le Lann M and Subias A 2002 Learning Methodology for a Supervision System using LAMDA Classification Method In IBERAMIA 02 VIII Iberoamerican Conference on Artificial Intelligence Sevilla Spain Kempowsky T Aguilar Martin J Subias A and Le Lann M 2003 Classification Tool based on Interactivity between Expertise and Self learning Techniques In IFAC Safeprocess Washington D C USA Kempowsky T 2004 Surveillance de proc d s base de m thodes de classification conception d un outil d aide pour la d tection et le diagnostic des d faillances Ph D Dissertation INSA 21
19. he user allows unclassified individuals meaning that an individual has not been recognized in any class its adequacy degree is lower than the minimum threshold and has been placed in the NIC class or force every individual to be assign to a class In this last case the Non Informative Class is not taken into account for recognition Figure 14 t Recognition Options x Do vou allow unclassified individuals FIGURE 14 RECOGNITION OPTIONS WINDOW 3 1 5 Data Processing Treatment and Visualization Once the context and the population have been loaded SALSA identifies the number of descriptors and its nature and it internally rescales quantitative values between o and 1 Normalisation process To make possible a direct confrontation between individuals and classes it is necessary that the descriptors used for observations are of the same type and in the same order than the concepts used for classes context However whenever the number of descriptors or its nature or the actual values exhibited in the population do not fit the ones stored in the context the problem is reported in an error file named error out The error is explained in this file with an indication of its nature type and location number of individual and number of descriptor if necessary A message window is displayed to notify the user If the population can still be loaded for instance if a value is lower than the minimum value
20. ll be displayed Class Profile Save Ref Cl View Prof CI Analyse FIGURE 21 CLASSIFICATION VISUALIZING BUTTONS CLASSIFY Class Details Figure 22 shows the profile for each class Each descriptor is represented with a different colour for quantitative descriptors their mean value is displayed for qualitative descriptors each modality is represented with a different colour and their occurrence frequency is shown the sum of all modalities for a qualitative descriptor must be 1 It is also possible to view the complete information for each class in text mode Class details button Figure 21 The Class Details window shows in a table all the parameters values for each class Also the number of elements individuals assigned to each class is displayed Figure 23 Depending on the type of algorithm used for classification different information will appear The user may select to view and export into a CSV file the class parameters either in a normalized format or with its absolute values 11 Class Profile 4 Class 1 Class 2 Current Classes Descriptor 2 CO 2 ValF Descriptors 02 E p ct D escriptor 4 O2 ValF Normalized Walues Absolute Values 20 8124 0 7053 19 0300 2 6459 20 1244 1 4543 5 9633 9 9902 4 1800 13 3702 Normalised values 7 Absolute values Export Print OF FIGURE 23 CLASS DETAILS WINDOW 3 1 6 2 Recognition Results When Recognition is launched
21. loaded the user must choose a classification file with CLA extension This classification must have the same number and type of descriptors as the loaded context Once the user has loaded the classification the learning parameters used for this classification will appear under the labels Presence Function Connectives and Exigency Index Once both files have been loaded the user may close the configuration window by clicking on the OK button 3 2 2 Online Connection Once the configuration window is closed the user will see the list of the quantitative descriptors or the qualitative descriptors that will be received online see Figure 35 LiUantitative Descriptors Peer rere etree teeters eee etter eee eee eee eee 04 OS alr DE T1_ alF DS T2_ValF D10 Ta_alF FIGURE 35 LIST OF ONLINE QUANTITATIVE DESCRIPTORS 17 View Configuration FIGURE 36 READY TO SEND ONLINE CONFIGURATION The Send Configuration button is enabled Figure 36 The user must click this button for SALSA to get ready to receive online data If the connection is successful SALSA will start checking for a new message to arrive Yew Configuration Send Configuration FIGURE 37 READY TO RECEIVE ONLINE DATA The Get Data button is enabled Figure 37 SALSA will display the path where the data transfer will take place Figure 38 The DataTr folder must contain the file with the current information message
22. means losing any previous classification FIGURE 6 LOADING A NEW POPULATION 6o Because of the presence of commented lines with a symbol and a possible supervised learning field there exist a couple of dialog windows that ask the user if the population file is to be looked at or edited with a Windows text editor and if the last field is to be treated like an imposed classification value See Figure 7 View Edit Population FIGURE 7 VIEW EDIT FILE amp IMPOSED CLASSIFICATION WINDOWS 3 1 2 Choosing the Classification Algorithm Quantitative descriptors To make the different calculations required in the learning and classification procedures it is necessary to choose the type of algorithm to be used for the quantitative descriptors Figure 8 Presence Function Lamdal FIGURE 8 PRESENCE FUNCTION SELECTION Lamda1 0 1 0 Lamda2 Ko 1 0 K logl 2 2p 1 Lamda3 p L P 1 x c I lt oef 2 Lamda4 Kp 1 0 op Ip o l pY u Gauss1 e Io Gauss2 Ke 20 with adjustable minimum threshold GAD NIC 3 1 3 Logic operators and the exigency level 3 1 3 1 Logic operators The user may choose between two different families of logic operators T norms amp dual T conorms functions in order to aggregate all the Marginal Adequacy Degrees of an individual to a class Figure 9 These operators are Product Probabilistic Sum T norm amp T conorm and the Minimum Maximum
23. n supervised learning i e the additional field correspond to the pre defined class Type of Data a descriptor is assumed to be quantitative if it starts with a number a point or a minus sign and qualitative otherwise The name or tag for each descriptor must be a line preceded by the amp sign and they should be also separated by a blank space or tab Pre defined classes the additional field for supervised learning will indicate the class where the individual is to be assigned The algorithm expects these numbers to appear in a semi ordered way that is class 1 has to appear at least once before class 2 there has to be at least one 2 before any 3 and so on When these numbers are greater than o they create a class The zero o means that the individual is part of a population that later will undergo pattern recognition for the moment no class is assigned to this element Figure 3 shows an example of the training file format for the case of supervised learning Notice that no name has to be specified for the class column in the tag line LEVEL DAT This individual file collects data from a real pilot plant system which consists of two tanks and a control system to manage the level of the second one This data base has been generated from three possible situations normal operation class one a partial obstruction between tanks class two and the emerge
24. n data set generates classes and assigns significant functional states to these classes by a constant dialogue with the expert The learning stage includes e Class Generation a clustering classifier with the capacity to perform a supervised or unsupervised learning procedure Interaction with the expert is necessary in order to tune up the classifier parameters e Mapping Classes to States a dialog tool with the expert for the assignment of classes or groups of classes to meaningful functional states The online stage is the recognition stage which will determine de functional state of the process from online data of sensors and actuators or other information from the process variables During plant operation SALSA generates the current functional state of the supervised process by using real time online data For further information concerning SALSA toolbox or LAMDA classification algorithm the user may refer to Kempowsky 2004 Aguilar Martin and L pez de Mantaras 1982 2 2 TECHNICAL DATA e This version of SALSA was programmed in plain ANSI C although the underlying philosophy is object oriented e SALSA is a standalone tool it can operate on its own without need of another software tool e The platform for the application s development was LabWindows CVI latest version used 8 e Graphic User Interface e Source Code programming o The software may be installed in Windows 98 2000 XP Vista 2 3 SALSA SETUP 1
25. n the population file _ CONTEXT New vew Seve lt NOne gt o POPULATION New E ee NPP AAPAN SAPNA hone FIGURE 4 NEW CONTEXT amp NEW POPULATION BUTTONS e Loading a new context The context file normally contains the number of descriptors and its type numeric or symbolic It may be induced from a population or plain text file with a DAT extension or retrieved from a binary file previously saved with a CON extension Figure 5 If the context has been induced from a population the application will demand the user if he wants also to load the corresponding population to be used for classification z New Context Do vou want bo induce a context From a population or retrieve a previously used one Retrieve FIGURE 5 LOADING A NEW CONTEXT INDUCE OR RETRIEVE e Loading anew population This file contains the individuals that will be used for training It must have the same number and type of descriptors as in the context and it may contain an additional field corresponding to the pre defined classes to be used for the supervised learning case A dialog window will appear asking the user if he wants to load the corresponding context Figure 6 If changes have previously been made e g a descriptor has been disabled to the context they will be lost if the user accepts loading the new context also Population loaded Do you want also to load the corresponding context REMEMBER that
26. ncy situation total obstruction at the exit of tank two class three Used descriptors are level in tank one level in tank two both from 0 to 10 the desired output or control input only High or Low the control signal also from 0 to 10 and the derivatives of first second level and control signal coded Negative Large Negative Small Zero Positive Small and Positive Large We do have a last field of imposed class amp LEVEL_T1 LEVEL_T2 OUTPUT CTRL_SIG DERIV_T1 DERIV_T2 DERIV_CONTROL 9 107939 3 021866 A 5 622508 D D C 1 9 312390 2 988349 A 5 940937 BCD 1 7 678884 1 027721 C 4 984652 ABD 1 9 994041 2 507010 A 4 953032 DAE 1 9 344999 2 015358 C 3481344 E EA 1 9 031239 3 020050 A 5 714987 DD BO 8 579471 1 061223 C 4 925020 DBD 2 8 196658 0 961353 C 6 249512 E DB 2 8 619834 0 909940 C 6 692107 ADBO 7 391809 0 987452 C 5 994978 E D B 2 8 945312 3 056641 A 5 458984 BD B 1 6 353343 5 844610 A 1 251475 EAC 1 9 928058 6 129644 A 1 275735 C C C 3 9 927319 6 129463 A 1 276346 D C C 3 5 955299 6 138647 A 1 276663 C C C 0 9 958092 6 142730 A 1 283172 D C C 3 7259859 0 976562 C 5 537109 DBC 1 FIGURE 3 EXAMPLE OF THE DATA DAT FILE 3 1 1 2 Loading a training file To load a file the user must select either the New Context or the New Population buttons Figure 4 Although it is possible to start with any of the two options it is recommended to first load the context data file and the
27. ng Results Classes amp Partition QualitY cceccccccecccccccccccorccccocoeaeae 7 Figure 13 Supervised Learning Options WINKOW ssscccsssscccsssscccsssscccssscccessscccessscccessccseeesccsensscseoees 8 Figure 14 Recognition Options WINGOW sii vieusdienter cess set csuranet outta E a S EAEE 8 Figure 15 Quantitative amp Qualitative Descriptor WINKOWS sssccccssssssccccssssscccccnsssscccceessssccsensesceeneees 9 Figure 16 Context Description WINKOW ssscccssssccccsssscccnssscccesscccessscccassscccassscccasscssnssceseesessaessessones 9 Figure 17 Active quantitative descriptors graph veccesecedscccecnesceesdedeecevctedsscdetducsecgiexcuavetdactonsbereustordeccorsaene 10 Figure 18 Active qualitative descriptors Sraph ccccsssscsccccssssscccccsssssccccesssseccceassssccsseesssccsseassseceseneees 10 Figure 19 Application ready fOr classification 2 cisccaecscscheecaaiaraece A A AS 10 Figure 20 Classification Results using Supervised Learning cccsessscccccsssssccccensssscccceessscccseassssceseneees 11 Figure 21 Classification visualizing buttons ssseeeesseeeecosoececosoececoseececososeeeosseeecossseeeossscecossseecosseeeeossseee 11 Figure 22 P FOMMETOl CACM Clas Ss ceaccesscraecees craccecaasainases cog avcncrasayen A 12 Figure 23 Class Details WINdOW sacccovieves cecexcctewlezeccveceneviunl cues ive ces ectays ARR cea sevecteylcaecceelenestanleseusexieseeaveenetes 12
28. ntitative Descriptors FIGURE 17 ACTIVE QUANTITATIVE DESCRIPTORS GRAPH o T ie T Descriptors Modalities TE Active Qualitative Descriptors Oo oh al 1 1 I I I I E q T3 E p Act 12 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 489 50 52 54 56 58 sft Active Qualitative Descriptors Individuals View Quantitative Descriptors FIGURE 18 ACTIVE QUALITATIVE DESCRIPTORS GRAPH 3 1 6 Classification Results and Visualization After successfully loading a context and population green LEDs for Context and Population and name of the loaded file it is possible to put the algorithm to work by pressing the CLASSIFY button see Figure 19 CONTEXT New View Save ATA_TEST Gazetieur _Episodes pprent_ 54 2D dat e POPULATION New View Sy aba ATA_TEST GazetieurEpisodes 4pprent_5S V2D dat CLASSIFICATION lt not saved Presence Function H Gauss Connectives Probabilistic F Exigency Index F 1 000000 Classification Mode Self Learning Class Profile Save Ref Cl Text CLASSIFY x Class Details Vie fiait D Analyse FIGURE 19 APPLICATION READY FOR CLASSIFICATION 10 2 OFFLINE SALSA Situation Assessment using LAMDA claSsification Algorithm C SYSTEM BUSY 3 CONTEXT New View Save ATA_TEST Gazeifieur_Episodes A4pprent_5 2D
29. on and online identification of the functional states of a process by means of a learning classification technique It combines qualitative and or quantitative information of the process variables and the expert s knowledge Kempowsky et al 2002 Kempowsky et al 2003 The manual is divided into two main chapters and an appendix as follows Chapter 2 Covers some technical information the installation of the application and the system requirements Chapter 3 This section provides sufficient information to use the application It shows an overview of the basic features of SALSA as well as the steps that must be followed Appendix Correspond to the installation of an online data emulator SALSA Online data r generator SALSA application as well as this user manual was developed by the DISCO group from the LAAS CNRS All rights reserved For further information and details please contact Joseph AGUILAR MARTIN aguilar laas fr Marie V ronique LE LANN mvlelann laas fr Tatiana KEMPOWSKY kempowsky laas fr This is a 2009 version 1 0 of SALSA User s Manual Updates are constantly made 2 SOFTWARE INSTALLATION 2 1 INTRODUCTION The main function of SALSA is to identify in a qualitative way the current functional state of a supervised process using the online symbolic and or numeric information from the process variables To do this SALSA has two main stages The first offline corresponds to a learning stage which from a give
30. r may choose to visualise either the quantitative or the qualitative data Abell 1 0 Normalized Values 1 6 10 15 20 25 30 35 40 50 55 60 65 70 75 80 85 90 9 101 QUANTITATIVE DESCRIPTORS Seconds 93 00 FIGURE 42 ONLINE QUANTITATIVE DESCRIPTORS E ng a L Dr m E La Ds pal 18 GGGGGGGGGCGGGGGGGGGGGGGGGGGEGGGGGGEGGAAAAAAFFFGGS piel l l l I I l l l l I I l l l l I l I l l l 1 F 10 15 20 25 30 35 40 45 mO 55 60 65 70 75 50 55 90 95 101 QUALITATIVE DESCRIPTORS Seconds 43 00 FIGURE 43 ONLINE QUALITATIVE DESCRIPTORS 3 2 3 2 Visualizing the Current State e The user may visualise the current state of the plant at the top of the screen a bar graph indicating the state of the process each time a new observation arrives Figure 44 19 CURRENT STATE 3 ALARM STATES I l l I I I l I I I l l I 1 5 10 15 20 25 30 35 40 45 50 55 GQ Gs r 75 20 25 g0 95 101 Seconds FIGURE 44 BAR GRAPH WITH STATES e A Table with the timestamp of a change of state the state and its associated class for the last 10 states is displayed The current state is highlighted with the corresponding colour of the state Figure 45 TIMESTAMP STATE _ CLASS JE D4 05 2006 16 23 06 3 ALARAM CL4 04 05 2006 16 22 48 5 RECOVER CL 04 05 2006 16 22 40 2 HIGH WATER es Be 04 05 2006 16 22 24 1 NORMAL OF CL FIGURE 45 TABLE WITH CHANGE OF STATE OCCURRENCES
31. s for context and classification 3 2 SALSA ONLINE ONLINE SALSA CURRENT STATE _ I 1 1 1 Start Time 100 110 120 130 140 150 160 170 180 F gt b Seconds Quantitative Descriptors 4 r t f D 2 T D D tw Me i 0 0 I I I 1 I I I I 1 I I I I 0 10 20 30 4 400 410 120 130 140 150 160 170 180 Graph Width 200 QUANTITATIVE DESCRIPTORS Seconds 0 00 TIMESTAMP GAD CLASS GAD NIC 0 0000E 0 4 OO00E 0 Recognition time 0 FIGURE 33 SALSA ONLINE MAIN SCREEN 16 These are the steps the user must follow for online recognition 3 2 1 Online Configuration The online configuration button is the only enabled button when the user chooses the online stage of SALSA in the main menu Figure 33 z gt SALSA Online Configuration Sampling Time sec Presence Function Landa Connectves Exigency Index FIGURE 34 SALSA ONLINE CONFIGURATION e The user may adjust the sampling time for the online recognition stage The possible values go from every 0 5 seconds up to every 5 minutes e The user must load up the context that will be used for recognition i e the number and type of descriptors used on the offline stage for the creation of classes Figure 34 This file is a binary file with a CON extension this file should have been previously created during the offline stage e Once the context is
32. sewecateesxpontess 18 Figure 38 PathiTor data TaN lE ae a ERE CRAC R aed teat CURA GE CRR AMNA T 18 Figure 39 Format TOF the online data iO wececkstccaccestcssestatcecctastcanechatcanecastenaeehataunecantussechateaeenteans 18 Figure 40 Format for the output file by SALSA eccccccecccccccccccccoccccososocco0ooooco0ooeoce0ooeocouooeoceuooocousesecenees 18 Figure 41 Stop receiving Ca taecrcscsesccctvedtancrducerscutiannndstiouwesanenendedawncnartdearndbbucempeatieameacnnesdeehs aa aT 19 Figure 42 Ontine Quantitative DCSEMDLONS si ccnstsienussatcuaccnc AA excouneanotoenvseniges maces 19 Figure 43 Onine Qualitative DESCHIDLOIS neoni eria Ge E GABE este TR ses ete ee esos 19 Figure 44 Bar Graph With States s lt civiescesetssstewswedesionndateveonsetiescndatacesedetsennnseonesgesons deatennetecwessedecuersannedeeess 20 Figure 45 Table with change of State occurrences cccecccccceccccccccccccccccococecoccseccccceococcccesoococeeuuus 20 Figure 46 Online Data Generator setup WINndOW ccccoceccccccccccccecccccccccccococcccccccccccecocccceeocccceeoooocveuuus 20 Figure 47 Online Data generator Main SCreeh ceeccccececccccccccccccccccccccccccccccccccecococceeooocceesooooeeueus 21 1 ABOUT THIS USER MANUAL This manual is a reference for the qualitative situation assessment of a given industrial process It provides the guidelines for the construction of classification systems for the offline characterisati
33. upervised learning has been selected are the maximum desired variation percentage Max Variability and the maximum allowed iterations number Max Iterations They are necessary because in unsupervised learning the class parameters vary so to obtain some stability and to overcome in a certain way the effect of the observations ordering the same population is classified several times until a maximum percentage of individuals changes from one iteration to another However the time to reach stability could be very long so a maximum number of iterations have also been introduced This means that the population will be presented to the existing classes once and again until either no more than the specified percentage of individuals varies its assignment from the previous iteration or the cycles limit has been reached In either case the total iterations made by the algorithm is displayed see Total Iterations in Figure 11 3 1 4 2 Active Supervised Learning The loaded population has a class imposition field This option allows performing a different number of choices like learning from an initial set of classes which can be modified by adding new classes or by updating their parameters or both Figure 13 Supervised Learning Options New Classes amp Class Modification Only New Class creation FIGURE 13 SUPERVISED LEARNING OPTIONS WINDOW 3 1 4 3 Pattern Recognition This option has two alternatives either t
Download Pdf Manuals
Related Search