Development of FPGA based Standalone Tunable Fuzzy Logic Controllers

Abstract

Soft computing techniques differ from conventional (hard) computing, in that unlike hard computing, it is tolerant of imprecision, uncertainty, partial truth, and approximation. In effect, the role model for soft computing is the human mind and its ability to address day-to-day problems. The principal constituents of Soft Computing (SC) are Fuzzy Logic (FL), Evolutionary Computation (EC), Machine Learning (ML) and Artificial Neural Networks (ANNs).
This thesis presents a generic hardware architecture for type-I and type-II standalone tunable Fuzzy Logic Controllers (FLCs) in Field Programmable Gate Array (FPGA). The designed FLC system can be remotely configured or tuned according to expert operated knowledge and deployed in different applications to replace traditional Proportional Integral Derivative (PID) controllers. This re-configurability is added as a feature to existing FLCs in literature. The FLC parameters which are needed for tuning purpose are mainly input range, output range, number of inputs, number of outputs, the parameters of the membership functions like slope and center points, and an If-Else rule base for the fuzzy inference process. Online tuning enables users to change these FLC parameters in real-time and eliminate repeated hardware programming whenever there is a need to change. Realization of these systems in real-time is difficult as the computational complexity increases exponentially with an increase in the number of inputs. Hence, the challenge lies in reducing the rule base significantly such that the inference time and the throughput time is perceivable for real-time applications.
To achieve these objectives, Modified Rule Active 2 Overlap Membership Function (MRA2-OMF), Modified Rule Active 3 Overlap Membership Function (MRA3-OMF), Modified Rule Active 4 Overlap Membership Function (MRA4-OMF), and Genetic Algorithm (GA) base rule optimization methods are proposed and implemented. These methods reduce the effective rules without compromising system accuracy and improve the cycle time in terms of Fuzzy Logic Inferences Per Second (FLIPS). In the proposed system architecture, the FLC is segmented into three independent modules, fuzzifier, inference engine with rule base, and defuzzifier.
Fuzzy systems employ fuzzifier to convert the real world crisp input into the fuzzy output. In type 2 fuzzy systems there are two fuzzifications happen simultaneously from upper and lower membership functions (UMF and LMF) with subtractions and divisions. Non-restoring, very high radix, and newton raphson approximation are most widely used division algorithms in hardware implementations. However, these prevalent methods have a cost of more latency. In order to overcome this problem, a successive approximation division algorithm based type 2 fuzzifier is introduced. It has been observed that successive approximation based fuzzifier computation is faster than the other type 2 fuzzifier.
A hardware-software co-design is established on Virtex 5 LX110T FPGA board. The MATLAB Graphical User Interface (GUI) acquires the fuzzy (type 1 or type 2) parameters from users and a Universal Asynchronous Receiver/Transmitter (UART) is dedicated to data communication between the hardware and the fuzzy toolbox. This GUI is provided to initiate control, input, rule transfer, and then to observe the crisp output on the computer. A proposed method which can support canonical fuzzy IF-THEN rules, which includes special cases of the fuzzy rule base is included in Digital Fuzzy Logic Controller (DFLC) architecture. For this purpose, a mealy state machine is incorporated into the design. The proposed FLCs are implemented on Xilinx Virtex-5 LX110T. DFLC peripheral integration with Micro-Blaze (MB) processor through Processor Logic Bus (PLB) is established for Intellectual Property (IP) core validation. The performance of the proposed systems are compared to Fuzzy Toolbox of MATLAB. Analysis of these designs is carried out by using Hardware-In-Loop (HIL) test to control various plant models in MATLAB/Simulink environments.