Login | Register

Machine Learning-based Strategies for Robust Fault Detection and Identification of Mobile Robots

Title:

Machine Learning-based Strategies for Robust Fault Detection and Identification of Mobile Robots

Baghernezhad, Farzad (2012) Machine Learning-based Strategies for Robust Fault Detection and Identification of Mobile Robots. Masters thesis, Concordia University.

[img]
Preview
Text (application/pdf)
Baghernezhad_MSc_S2013.pdf - Accepted Version
26MB

Abstract

Nowadays, process monitoring and fault diagnosis techniques are becoming a critical component of modern automatic control systems. One of the most crucial issues for the design of automatic control systems is reliability and dependability. Traditional ways to achieve these goals are through designing adaptive and robust controllers to eliminate any influence of faults on an output. Using these approaches, faults are managed but they could ultimately lead to failures; after which no controller could repair such effects. In order to minimize such damages, it is necessary to diagnose and rectify faults as soon as possible. In a fault detection system, residual generation is the first step in detecting faults, but residuals are not the only element of a dependable fault detection system. A fault detection system is reliable when an appropriate residual evaluation method is used along with a suitable residual generation technique.

The problem of fault detection and identification in a nonlinear system with applications to mobile robots is addressed in this thesis. For this purpose, first a new simulator for mobile robots containing kinematic and dynamic equations of a mobile robot and its actuators is designed. To detect faults in the system, a linear velocity of the mobile robot is chosen and modeled with computationally intelligent techniques. Locally linear models (LLM) as a neuro-fuzzy technique and radial basis function (RBF) as a powerful neural network are used to estimate the linear velocity of a mobile robot and generate residuals by comparing these with the system measurements. Subsequently, residuals are evaluated by using fixed and adaptive threshold bands. Adaptive threshold bands are generated using locally model thresholds (LMT) and model error modeling (MEM) technique in order to reduce the fault detection delay and false alarms. Finally, fault identification of a mobile robot by using multiple model technique is presented with the two proposed methods of modeling and threshold generation. The fault identification task consists of determining the occurrence of the fault as well as its location and magnitude. This is accomplished on four different types of faults with different magnitudes that are divided in ten different classes. For each scenario, simulation results are presented to demonstrate and illustrate the advantages and disadvantages of each methodology.

The main contributions of this thesis can be stated as follows: (a) the development and design of a new fault diagnosis method and a residual evaluation scheme by using an adaptive threshold band that is accomplished by using locally linear models of the system, (b) development of a fault detection approach based on computational intelligent algorithms for mobile robots using the MEM algorithm to generate adaptive threshold bands, (c) development of two fault identification approaches based on the concept of multiple models for a mobile robot, and (d) the resulting improvements of the proposed adaptive threshold bands are shown through extensive simulation results.

Divisions:Concordia University > Gina Cody School of Engineering and Computer Science > Electrical and Computer Engineering
Item Type:Thesis (Masters)
Authors:Baghernezhad, Farzad
Institution:Concordia University
Degree Name:M.A. Sc.
Program:Electrical and Computer Engineering
Date:5 December 2012
Thesis Supervisor(s):Khorasani, Khashayar
ID Code:974992
Deposited By: FARZAD BAGHERNEZHAD
Deposited On:06 Jun 2013 19:16
Last Modified:18 Jan 2018 17:39
All items in Spectrum are protected by copyright, with all rights reserved. The use of items is governed by Spectrum's terms of access.

Repository Staff Only: item control page

Downloads per month over past year

Back to top Back to top