Controlled Slip Design with NeuFuz
Contents
1. 1995 National Semiconductor Corporation TL DD12318 RRD B30M75 Printed in U S A zn4n N YUM UBISEq IIS p llo1u09 696 NV NeuFuz4 NeuFuz4 is a software tool to automatically generate a con trol program for the COP8 microcontroller The control pro gram uses fuzzy logic concepts the rules and membership functions for the fuzzy logic equations are generated auto matically using the back propagation learning algorithm The rules and membership functions are used to implement any general non linear function of between one and four vari ables Given several representative examples of the desired input and output data points NeuFuz4 will learn a general solution From the general solution NeuFuz4 will generate a source code program in COP8 assembly language that can be used to implement the control function The data used by NeuFuz4 to generate a solution can come from many different sources The four most common sourc es are a manually entered concept the output of a simula tion a theoretical model or collected from an existing appli cation A manually entered concept requires knowledge of the application The knowledge can be either fuzzy in con cept in which case NeuFuz4 offers A Better Fuzzy Logic than Fuzzy Logic since NeuFuz4 can generate the re quired rules and membership functions knowing only the center values of the desired results For example a set of fuzzy rules that can be used to generate t2. Controlled Slip Design with NeuFuz FEATURES m Provides accurate slip control based on NeuFuz algo rithm Reduces design time Requires minimum hardware Can be used to duplicate existing control solutions Can be used for traction control or ABS systems Firmware extremely modular Firmware for new wheel type is possible without writing any microcontroller code m Solution valid under varying friction conditions INTRODUCTION This application note focuses on a design based on Nation al Semiconductor s patented NeuFuz based neural fuzzy technology to arrive at a slip control scheme implemented on National Semiconductor s low cost COP8 microcon troller NeuFuz technology allows for the automated genera tion of a fuzzy logic control engine based on training data In this case the training data was taken from the well known characteristics of controlled slip Using the NeuFuz4 code generator the NeuFuz based design methodology followed here allows for tailoring the controller to suit different motor and wheel characteristics without having to rewrite the code Drive wheel slip is defined as the difference in the tangential velocity of a rotational drive wheel and the surface being driven In vehicular drive applications slip is the difference in wheel speed tangential tire velocity and vehicle speed speed of the vehicle Controlling the magnitude of slip is the basis of vehicular applications such as Anti lock Brake S
3. any desired control func tion The development time for the slip control program us ing the NeuFuz4 code generator was four days for a com plete solution Firmware Modules The firmware for the slip controller is extremely modular It can therefore be tailored to design a system with a mix and match of needed features The complete neural fuzzy development cycle to create the slip control algorithm consists of the following steps Define the data set Preprocess the data set to tailor it to the hardware Configure the NeuFuz neural network Train the neural network Find an optimized fuzzy representation Generate code Integrate code with other code in the target system Data Set Definition The first step in a NeuFuz design is to decide on the control input parameters In order to avoid data scaling problems the data set is selected so that 8 bit values read directly from the speed counters are used as input to the FUZZ routine One of the five bytes of OUT is used to directly load the PWM counter If No Data Exists The best use of NeuFuz is the generation of high precision solutions from desired performance data In the case of the slip control application the desired performance is a well known control surface Many applications exist where such information is not available If very little information is known about the system to be controlled and very little data is available the flexibil
4. OMPONENTS IN LIFE SUPPORT DEVICES OR SYSTEMS WITHOUT THE EXPRESS WRITTEN APPROVAL OF THE PRESIDENT OF NATIONAL SEMICONDUCTOR CORPORATION As used herein 1 Life support devices or systems are devices or systems which a are intended for surgical implant into the body or b support or sustain life and whose failure to perform when properly used in accordance with instructions for use provided in the labeling can be reasonably expected to result in a significant injury to the user 2 A critical component is any component of a life support device or system whose failure to perform can be reasonably expected to cause the failure of the life support device or system or to affect its safety or effectiveness National Semiconductor Japan Ltd Sumitomo Chemical Engineering Center National Semiconductor GmbH Livry Gargan Str 10 D 82256 Furstenfeldbruck National Semiconductor Corporation 2900 Semiconductor Drive P O Box 58090 N Santa Clara CA 95052 8090 Germany Bldg 7F Tel 1 800 272 9959 Tel 81 41 35 0 1 7 1 Nakase Mihama Ku TWX 910 339 9240 Telex 527649 Chiba City Fax 81 41 35 1 Ciba Prefecture 261 Tel 043 299 2300 Fax 043 299 2500 National Semiconductor Hong Kong Ltd 13th Floor Straight Block Ocean Centre 5 Canton Rd Tsimshatsui Kowloon Hong Kong Tel 852 2737 1600 Fax 852 2736 9960 National Semiconductor Australia Pty Ltd Building 16 Business Park Driv
5. a file can be written directly as Better Fuzzy than Fuzzy Data for Traction Control Wheel Speed Vehicle Speed APWM Counter 0 0 127 127 0 0 255 0 127 0 128 0 127 128 127 255 128 0 0 255 127 127 255 0 255 255 127 From this very sparse data NeuFuz4 can generate a com plete fuzzy logic solution While conventional fuzzy logic re quires extensive guess work to determine the width and height of membership functions NeuFuz4 can interpolate between values to give a solution Preprocess the Data Set Once the control input and output parameters are known a table containing the values and corresponding output are made It is recommended that sufficient data points are available to account for the nonlinearities of the system The data points must span all the possibilities of input val ues within the input space The table must be in the form of an ASCII file The NeuFuz4 user manual provides useful information on preparing the ASCII file Configure the Neural Net The configuration parameters for training the neural net the number of fuzzy membership functions desired and the ab solute accuracy desired from the system need to be de fined Train the Neural Network Training the neural network is an iterative process This re quires the user to study the error generated during training and to modify the learning neural network s parameters when needed The NeuFuz4 training for both the ABS and tractio
6. division and can be shared by other firmware modules Integrate Code with Other Firmware Modules In this application the unscaled input is an 8 bit value and is stored in RAM locations IN1 and IN2 The fuzzy logic algo rithm reads data from these RAM locations and writes the output in RAM locations labeled OUT1 to OUTS One of the most significant benefits of using the fuzzy logic assembly code produced by NeuFuz4 is that the RAM used by it can be reused by other assembly modules Should the NeuFuz4 generated code be interrupted during execution it is necessary to protect all the contents of RAM used by NeuFuz4 Special care must be taken not to over write the RAM locations that NeuFuz4 uses NeuFuz Design Implications Neural networks and fuzzy logic are highly suitable for mod eling non linear time variant system behavior Conventional linear control can only perform a linear approximation of a nonlinear behavior This approximation may be sufficient for some applications but not suitable for all especially when a high degree of accuracy is desired Neural networks and fuzzy logic have proven to be highly suitable for such appli cations Although these two technologies individually suffer from certain drawbacks when combined as in NeuFuz these disadvantages can be successfully eliminated main taining all the advantages NeuFuz allows the designer to take advantage of the learning capability of neural networks at the same t
7. e Monash Business Park Nottinghill Melbourne Victoria 3168 Australia Tel 3 558 9999 Fax 3 558 9998 National Semiconductores Do Brazil Ltda Rue Deputado Lacorda Franco 120 3A Sao Paulo SP Brazil 05418 000 Tel 55 11 212 5066 Telex 391 1131931 NSBR BR Fax 55 11 212 1181 National does not assume any responsibility for use of any circuitry described no circuit patent licenses are implied and National reserves the right at any time without notice to change said circuitry and specifications
8. ime providing a cost effective fuzzy logic imple mentation of the system It offers a high level of automation in the design process and significantly reduces design time It allows the designer to concentrate on the system configu ration and performance while hiding all the error prone cumbersome mathematical manipulations It provides more control over the design by introducing an added feature to specify the accuracy of the fuzzy system as well as better modeling of nonlinear behavior The result is improvement in performance and reduction in cost the advantages of NeuFuz based design make it a clear choice for microcon troller based slip control ith NeuFuz ign wi Controlled Slip Des AN 969 Results The COP8 slip controller is able to perform both the ABS and traction control functions better than a human operator The slip control circuit can be completely implemented with two chips a COP884CG microcontroller and an LM12298 H bridge By using NeuFuz4 an accurate reliable high per LIFE SUPPORT POLICY formance solution control program was generated for the COP8 in a much shorter time than would be required for a manually written program NeuFuz4 is extremely flexible training data may be either precise values for high perform ance solutions to well known problems or rough fuzzy logic type approximations if not precise training data is available NATIONAL S PRODUCTS ARE NOT AUTHORIZED FOR USE AS CRITICAL C
9. ity of NeuFuz4 allows it to be used as a better form of fuzzy logic For example in the case of the traction case of the slip control application the rules can be intuitively stated as follows IF wheel speed IS small AND vehicle speed IS small THEN slip IS small SO increase speed IF wheel speed IS medium AND vehicle speed IS small THEN slip IS medium SO hold speed IF wheel speed IS large AND vehicle speed IS small THEN slip IS large SO decrease speed IF wheel speed IS small AND vehicle speed IS medium THEN slip IS medium SO hold speed IF wheel speed IS medium AND vehicle speed IS medium THEN slip IS small SO THEN increase speed IF wheel speed IS large AND vehicle speed IS medium THEN slip IS medium SO hold speed IF wheel speed IS small AND vehicle speed IS large THEN slip IS large SO decrease speed IF wheel speed IS medium AND vehicle speed IS large THEN slip IS medium SO hold speed IF wheel speed IS large AND vehicle speed IS large THEN slip IS small SO THEN increase speed Conventional fuzzy logic systems require that the member ship functions for large medium and small be fully defined NeuFuz4 requires only that an approximate value close to the center value of the membership functions be declared The width height and location of each membership func tion and the values of the rules will be automatically deter mined by the Neural Network in NeuFuz4 For the rules giv en above for example the dat
10. n control surfaces has 440 data points each Using a learning rate of 0 1 and a learning factor of 0 01 the neural network will converge to an epsilon of 0 5 in less than 50 hours on a 486DX50 Since both the inputs and the output range be tween 0 and 255 an epsilon of 0 5 gives 1 2 LSB accuracy Convergence time is strongly dependent on the number of data points in the training data For example the rough solu tion given by the Better Fuzzy then Fuzzy data with only nine data points will converge in a few minutes on a 486DX50 Find an Optimized Fuzzy Representation The fuzzy logic solution obtained from the trained neural network needs to be verified for accuracy and size The accuracy of the solution is verified over the entire range of input space This fuzzy logic solution can be further opti mized directly from NeuFuz4 using a deletion factor to elimi nate some of the less significant rules with minimal effect on the accuracy of the solution Generate Code Once the neural network has been trained and the accuracy of the fuzzy logic solution found acceptable NeuFuz4 auto matically generates COP8 code The code generated by NeuFuz4 comprises of relocatable COP8 assembly code The code generated also includes the definitions for the RAM requirements A log file indicating the amount of RAM and ROM used for this algorithm is also generated The COP8 code includes some general purpose math routines for multiplication and
11. raining data for the slip application can be written intuitively A manually entered concept may also be a control surface which is well known Both the traction control and the ABS applications have well known control surfaces The control surfaces from available data tables give the best results and were selected for this application Data collected by observation from simulations or theoretical models may be used to generate data that can be used for a final solution Data collected from existing solutions may be used to reverse engineer control solutions Programs generated by NeuFuz4 will be predictable and re liable Since all programs generated by NeuFuz4 are essen tially identical except for tables of fuzzy logic parameters the reliability of the code is assured The control function can be changed by changing the fuzzy evaluation table and leaving the main fuzzy inference program unchanged Programs generated by NeuFuz4 have a high level of inte gration Since NeuFuz4 generates a program that maps all byte combinations of between one and four inputs to a five byte output the need for preprocessing of data before a control algorithm is executed is drastically reduced For the slip control application the entire algorithm was implement ed in NeuFuz4 so that no preprocessing was required Generating programs using NeuFuz4 shorten development time since NeuFuz4 automatically implements in a reliable and execution efficient program
12. re 8 bit unsigned integers read from a counter that counts timing pulses for the two wheels The output speed value is an 8 bit unsigned integer written to a counter that generates a PWM motor drive signal The smaller counter values represent greater speed COP8 Microcontroller The slip control application was implemented using a Na tional Semiconductor COP884CG The COP884CG is one of National Semiconductor s family of fully static 8 bit CMOS microcontroller s built around the common COP8 core All COP8 microcontrollers are capable of executing the same NeuFuz4 code For any application the particular COP8 that offers the optimum feature set can be selected For example the slip control problem requires that two opti cal timing disks be monitored and a PWM motor control signal be generated The COP884CG has three 16 bit coun ter times For the slip control application two counter timers are used to count timing wheel pulses and the third counter timer is used to generate a PWM control signal For other applications which require analog inputs the COP884CF trades off one 16 bit counter timer for an 8 bit SAR type A D converter NeuFuz4 and the appropriate COP8 can handle data from any source provided that the inputs can be expressed as one byte values Microcontroller COP 880 TL DD 12318 1 FIGURE 1 NeuFuz Based Slip Controller Block Diagram COP8 and NeuFuz are trademarks of National Semiconductor Corporation
13. ystems ABS and controlled traction transmissions Rath er than eliminate slip totally slip control systems are de signed to provide some slip to the system Typical automo tive ABS systems attempt to control slip so that a constant 20 slip rate is maintained during ABS operation The ex perimental system was constructed to investigate the slip control problem The problem was posed as a drive wheel Mechanical Coupling Slip Line Flywheel Drive Wheel Timing Pulses Flywheel Timing Pulses National Semiconductor Application Note 969 Dr W Shields Neeley September 1994 turning a flywheel as shown in Figure 1 The mechanical coupling between the drive wheel and flywheel was inten tionally kept loose If the drive wheel attempted to change the speed of the flywheel too quickly the high inertia of the flywheel would cause it to remain at its current rotational velocity In this case the condition of the flywheel following the speed changes of the drive wheel too slowly is called slip In this experiment traction control attempted to accel erate the flywheel to a predetermined velocity while ABS braking was accomplished by reversing motor direction and attempting to reduce the flywheel speed to a lower set point As a control problem two inputs and one output are needed for the experimental slip system The inputs are the drive wheel speed and the vehicle speed The input speed values a
Download Pdf Manuals
Related Search