Reconfigurable Fuzzy Automaton for Software Agents Janos L
Reconfigurable Fuzzy Automaton for Software Agents Janos L. Grantner, Paolo A. Tamayo, Ramakrishna Gottipati, George A. Fodor Presentation By Dr. Janos L. Grantner WCCI/FUZZ-IEEE 06
Presentation Outline • • • Introduction HFB-FSM Model Intelligent Software Agents Reconfigurable Architecture Design Simulation Results Conclusion 2
Introduction 1. 2. 3. 4. The problems that characterize industrial process control innovation are: Introducing new knowledge into a system Activating stored domain knowledge in an autonomous way Validating the knowledge Recovering the system if the new, activated knowledge is not suitable to handle the situation 3
Introduction (contn’d) These problems can be addressed using intelligent software agents with fuzzy automata • New knowledge can be implemented by adding agents – New knowledge is introduced by means of states in the goal path of an event driven, sequential control algorithm – Fuzzy automata is an effective approximation method to model continuous and discrete signals in a single theoretical framework • Knowledge validation is achieved – By quantifying the degree of deviation from the nominal operating conditions due to unexpected events – Execution monitoring is also performed with fuzzy automata 4
Intelligent Software Agent ARCHITECTURE To FSA-BROKER: architecture, Supervision, real-time To ALARM SERVER To HMI Server All ports are bi-directional Object State All ports have a named type Application Commissioning Panel Fuzzy Automaton Connection to other objects APPLICATION 5
Hardware Implementation of Agents • IP (Intellectual Property) modules are designed as generic fuzzy automaton agents • Agents communicate via No. C (Network on Chip) to decrease the real estate needed for pathways on the chip • Agent broker can be implemented on an FPGA • A set of specialized architecture operations are needed to implement an agent broker on an FPGA • Example of such implementation: No. C 6
Synthesis of Network on Chip (No. C) • Input: IP components with cost figures: U 1, U 2, U 3, V 1, V 2, V 3 • Clustered constraints (clustering is NP-complete): U 1, 2, 3 is one cluster, V 1, 2, 3 is another cluster • Communication Constrained Graph: Ui communicates with Vi, i=1, 2, 3 • Optimal synthesis – quadratic programming approach: have only one communication channel • Method: CDCS (constrained driven communication synthesis) • At present, software implementation takes minutes u 1 u 2 u 3 v 1 v 2 v 3 7
HFB-FSM Model 8
Example for Designing a Reconfigurable Fuzzy Automaton • It is based upon the computational model of HFB -FSM • Assumes a multi-fuzzy input and one fuzzy output (MISO) configuration • Digital inputs and analog inputs with threshold are omitted at this point • Each fuzzy input is mapped to a set of Boolean variables using the B-Algorithm (Fuzzy-to. Boolean mapping) 9
Example (Contn’d) k overlapping linguistic sub – intervals are mapped to n (n = 2 k-1) non overlapping Boolean sub – intervals, and Xbi = 1 if the xc position of the fuzzy input maximum falls into Boolean sub – interval i(i = 1, …, n) and XBj = 0 for all j = i(j = 1, ……, n) 10
Example (Contn’d) 11
Reconfigurable Architecture Design (Contn’d) 12
Example (Contn’d) 13
Example (Contn’d) 14
Example (Contn’d) 15
Example (Contn’d) 16
Example (Contn’d)-Parameters Component Description Number of Fuzzy Inputs Defines the number of fuzzy inputs of the system. This value also determines the number of parallel MOM and Interval Detection circuits. Resolution The resolution of the degree of membership is resizable. This determines the number of bits needed to represent the degree of membership. Granularity This resizable property depends upon the number of elements in the universal set. Boundary Count This is the number of Boolean sub-intervals. It can be reset from problem to problem. Boundary Limits The right-most element of each Boolean subinterval. The total number of limits is equal to the Boundary Count. Number of Fuzzy States The number of fuzzy states in the particular state cluster to be implemented. 17
Validation Container Crane Problem • The Container-Crane problem simulates the operation of transferring a container van from a ship into a railcar platform. The Container-Crane problem is developed using the fuzzy. Tech software 18
Validation (Contn’d) Container Crane Problem • The two fuzzy inputs are – Angle of displacement of the suspended load (X) • Left swing results in a negative angle • Right swing results in a positive angle – Distance of the load from the rail car (Y) • Far , Near , Close (also the states of the system) • Output is the power applied to the crane (Z) – Positive power, negative power and zero power • A simplified HFB-FSM will have 3 states, each of which will be made of just one crisp state 19
Validation (Contn’d) Rules Rule Description For Crisp State 1 (far) Rule 1 If X is zero and Y is far then Z is positive. Rule 2 If X is negative and Y is far then Z is positive. Rule 3 If X is positive and Y is far then Z is positive. For Crisp State 2 (near) Rule 1 If X is zero and Y is near then Z is positive. Rule 2 If X is negative and Y is near then Z is positive. Rule 3 If X is positive and Y is near then Z is negative. Rule 4 If X is zero and Y is close then Z is zero. Rule 5 If X is positive and Y is close then Z is negative. Rule 6 If X is negative and Y is close then Z is zero. For Crisp State 3 (close) Rule 1 If X is zero and Y is close then Z is zero. Rule 2 If X is positive and Y is close then Z is negative. Rule 3 If X is negative and Y is close then Z is zero. 20
Validation (Contn’d) Normalized universal set 21
Validation (Contn’d) Present State Next State Fuzzy Input (Angle) Fuzzy Input (Distance) Fuzzy Output (Power) Defuzzified Value 1 1 Zero Far [0. 0 0. 5 1. 0] 7 1 1 Negative Far [0. 0 0. 5 1. 0] 7 1 1 Positive Far [0. 0 0. 5 1. 0] 7 1 2 Negative Near [0. 0 0. 5] 6 2 2 Zero Near [0. 0 0. 5 1. 0] 7 2 2 Negative Near [0. 0 0. 5 1. 0] 7 2 2 Positive Near [1. 0 0. 5 0. 5] 2 2 3 Zero Close [0. 5 1. 0 0. 5] 4 3 3 Positive Close [1. 0 0. 5 0. 0] 2 3 3 Negative Close [0. 0 0. 5 1. 0 0. 5 0. 0] 4 22
Validation (Contn’d) Simulation 23
Validation (Contn’d) Inference and Model Building Operation Performance Summary Type of Operation Number Size of 1 Number of of Rules Fuzzy Elements in the per State Rule Universal Set Number of Clock Cycles Needed Inference R Nx. N N N+K Model Building R Nx. N N (Sx. Nx. R) + 1+K Where : K is the constant overhead cycles when performing the operation, currently 4 clock cycles. S is the number of States 24
Validation (Contn’d) For the example: • Inference will be N+K = 7 + 4 = 11 clock cycles. Model Building will be (Sx. Nx. R) + 1+K = (3 x 7 x 7 x 3) + 1+4 = 446 clock cycles • At 100 MHZ clock rate we can run approximately 220, 000 Model Building Operations and 10 Million Inferences per second 25
Conclusion • An intelligent software agent architecture with fuzzy automaton was introduced • Online reconfiguration of this architecture is needed to introduce new knowledge and for fault detection and identification and recovery • IP (Intellectual property) modules are implemented on hardware in contemporary control systems • Hardware implementation of a reconfigurable fuzzy automaton was presented. 26
- Slides: 26