November 22, 2024
Amin Torabi Jahromi

Amin Torabi Jahromi

Academic Rank: Assistant professor
Address:
Degree: Ph.D in Electrical Engineering
Phone: 09171023389
Faculty: Faculty of Intelligent Systems and Data Science

Research

Title
Fault Tolerant Dynamic Fuzzy Neural Network Based Control for a Nonlinear Process Using Algorithm Based Fault Tolerance
Type Thesis
Keywords
Algorithm Based Fault Tolerance مقاوم سازي الگوريتم در برابر نقص شبكه عصبي فازي ديناميكي - تحمل نقص - يادگيري خودسازماندهي آنالين - سيستم فازي TSK - كنترل كننده - افزونگي نرم افزاري - افزونگي محاسباتي - مقاوم سازي در برابر نقص مبتني بر الگوريتم
Researchers neda mahmoodi (Student) , Amin Torabi Jahromi (Primary advisor) , Valiollah Ghaffari (Primary advisor)

Abstract

Background: In this research, the control algorithms has been developed and proposed based on a dynamic fuzzy neural network which is tolerant to the faults of matrix calculations. The aim of this control algorithm is speed learning and more compact network structures and its high generalizable performance, improves the quality of the control. Thus, its tolerance to fault is of particular importance. Aim: The main aim of this research is to present an algorithm that implements faulttolerant dynamic fuzzy neural network (DFNN). This control algorithm, inspired by the direct inverse control method, uses the DFNN learning algorithm as an estimator and controller, which makes the system tolerant to faults and improves the control performance. The DFNN estimator and controller has been built on the basis of extended RBF, and it performs as a TSK system, which results in speed learning and a more compact network structure with high approximation and generalization performance, and is tolerant to damage caused by fault with a fault detection algorithm through data redundancy. In other words, it examines algorithm base fault tolerance (ABFT), which is a low-cost, high-performance scheme to detect and correct permanent and temporary errors in matrix operations. In the end, a non-linear process will be investigated in this approach and its efficiency can be observed in performing control operations. Each algorithm has its own particular execution time, which is one of the important issues of algorithm design, and its efficiency are checked according to its execution time. The execution time of the fault-tolerant algorithm is different with and without the fault and the execution time and efficiency of both are investigated during performing of the algorithms. Methodology: According to the DFNN learning algorithm, RBF units are first determined by entering the training data. To do this, criteria of neuron generation for hierarchical learning and factor of neuron generation, in