AFM 1.0 ADAPTIVE FUZZY MODELLER ADVANCED DATA Up to 8 Input Variables and 4 Output Variables Up to 8 Fuzzy Sets for each Input Variables Up to 214 Fuzzy Rules Rules Learning Phase using an unsupervised WTA-FAM Membership Functions Learning Phase using a supervised BACK-FAM Automatic and Manual Learning Rate Rules Minimizer Gaussian and Triangular Membership Functions Shape Inference method based on Product or Minimum Step-by-Step and from File Simulation available Supported Target: W.A.R.P. 1.1, W.A.R.P. 2.0, MATLAB and ANSI C DESCRIPTION Adaptive Fuzzy Modeller (AFM) is a tool that easily allows to obtain a model of a system based on Fuzzy Logic data structure, starting from the sampling of a process/function expressed in terms of Input\Output values pairs (patterns). Its primary capability is the automatic generation of a database containing the inference rules and the parameters describing the membership functions. The generated Fuzzy Logic knowledge base represents an optimized approximation of the process/function provided as input. The AFM has the capability to translate its project files to FUZZYSTUDIO project files, MATLAB and C code, in order to use this environment as a support for simulation and control . The block diagram in fig.2 illustrates the AFM logic flow. Figure 1. Block Diagram May 1996 This is advance information on a new product now in development or undergoing evaluation. Details are subject to change without notice. 1/4 ADAPTIVE FUZZY MODELLER 1.0 LEARNING it is composed by two phases: BUILDING RULES It allows to perform the automatic selection of inference rules or their manual definition, taking in to account the project constrains read from the previously opened pattern file. As a result the user will be supplied with a rule file containing the linguistic expression of the rules. An unsupervised clustering algorithm is used to perform this task. BUILDING MEMBERSHIP FUNCTIONS It allows the user to select the membership function shape and the fuzzy intference method for the project elaboration. Starting from the rule file supplied by the previous phase, it initially associates to each fuzzy set a standard membership function shape. These shapes can be gradually tuned in order to let the fuzzy system to better approximate the process/function sampling by means of subsequently run sessions. Back-propagation algorithm with automatic learning rate control is used to this aim. Figure 3. BUILD MEMBERSHIP FUNCTION window TOOLS It is composed of different sub-menus: LOCAL RULES it allows to add new rules to the fuzzy logic knowledge base determined by an Adaptive Fuzzy Modeller run session. Aim of this functionality is the local approximation level improvement. SIMULATION it allows to simulate the fuzzy system behaviour in order to verify the approximation level obtained during the learning phase. The simulation can be carried out in two different ways. Simulation Step-by-Step: the user must supply the simulator with the values variables corresponding to the point to verify. Simulation from File: the user must supply the simulator with the name of a process/function stream file that will be used to perform a complete process inference. 2/4 Figure 2. AFM Logic Flow. pattern file rules minimizer Learning Phases Fuzzy Logic knowledge base exporter to processor Rules extractor MFs tuning Simulation and Manual Tuning W.A.R.P. 1.1 W.A.R.P. 2.0 ANSI C MATLAB VIEW FEATURES View Features of the AFM gives with the capability to visualize the fuzzy model extracted for a particular project. It allows a separate visualization of the rules of inference and membership functions. The rules can be visualized in a linguistic format. For the membership functions you can choose between a linguistic and a graphical format visualization. EXPORTERS The Exporter provides library functions working on the databases automatically generated, which appropriately describe the data structures of the selected project in terms of a different programming environment. These functions can be exploited inside the user’s programs in order to verify the model extracted and to use it in real application. SUPPORTED TARGETS The supported environment are: - W.A.R.P.1.1 using FUZZYSTUDIO1.0 - W.A.R.P.2.0 using FUZZYSTUDIO2.0 - MATLAB - C Language - Fu.L.L. (Fuzzy Logic Language). ADAPTIVE FUZZY MODELLER 1.0 SYSTEM REQUIREMENTS MS-DOS version 3.1or higher Microsoft Windows 3.0 or compatible later version 486, PENTIUM compatible processor chip 8 MBytes RAM (16 Mbytes recommended) Hard Disk with at least 1MBytes free space Order Code Description Supported Target STFLWARP11/PG STFLWARP11/PL WTA-FAM for Building Rules STFLAFM10/SW STFLWARP20/PL BACK-FAM for Building MFs ANSI C MATLAB® Functionalities System Requirement Rules Minimizer MS-DOS 3.1or higher Step-by-Step Simulation Windows 3.0 or later Simulation from File 486, PENTIUM compatible Local Tuning 8 MB RAM Type Description Operating Temperature Package STFLWARP11/PG-PL HCMOS, 6KBytes RAM, 40MHz, 16 Inputs, 16 Outputs, 256 Rules 0 - 70°C CPGA100-PLCC84 STFLWARP20/PL HCMOS, 1.4KBytes RAM, 40MHz, 8 Inputs, 4 Outputs, 256 Rules 0 - 70°C PLCC68 Type Device STFLSTUDIO10/KIT STFLWARP11/PG STFLSTUDIO2/KIT STFLWARP20/PL Development Tools FUZZYSTUDIO ADB W.A.R.P. 1.X W.A.R.P. 1.X programmer EPROM programmer RS-232 communication handler Internal Clock W.A.R.P.2.0 W.A.R.P.2.0 programmer ZEROPOWER programmer RS-232 communication handler Internal Clock FUZZYSTUDIO SDT Variables and Rules Editor W.A.R.P. Compiler/Debugger Exporter for ANSI C and MATLAB® Variables and Rules Editor W.A.R.P. Compiler/Debugger Exporter for ANSI C and MATLAB® 3/4 ADAPTIVE FUZZY MODELLER 1.0 Information furnished is believed to be accurate and reliable. However, SGS-THOMSON Microelectronics assumes no responsibility for the consequences of use of such information nor for any infringement of patents or other rights of third parties which may result from its use. No license is granted by implication or otherwise under any patent or patent rights of SGS-THOMSON Microelectronics. Specification mentioned in this publication are subject to change without notice. This publication supersedes and replaces all information previously supplied. SGS-THOMSON Microelectronics products are not authorized for use as critical components in life support devices or systems without express written approval of SGS-THOMSON Microelectronics. © 1996 SGS-THOMSON Microelectronics – Printed in Italy – All Rights Reserved FUZZYSTUDIO is a trademark of SGS-THOMSON Microelectronics MS-DOS®, Microsoft® and Microsoft Windows® are registered trademarks of Microsoft Corporation. MATLAB® is a registered trademark of Mathworks Inc. SGS-THOMSON Microelectronics GROUP OF COMPANIES Australia - Brazil - Canada - China - France - Germany - Hong Kong - Italy - Japan - Korea - Malaysia - Malta - Morocco - The Netherlands Singapore - Spain - Sweden - Switzerland - Taiwan - Thailand - United Kingdom - U.S.A. 4/4