Login | Register

Fault Detection, Isolation and Identification of Autonomous Underwater Vehicles Using Dynamic Neural Networks and Genetic Algorithms

Title:

Fault Detection, Isolation and Identification of Autonomous Underwater Vehicles Using Dynamic Neural Networks and Genetic Algorithms

Shahrokhi Tehrani, Shaghayegh (2015) Fault Detection, Isolation and Identification of Autonomous Underwater Vehicles Using Dynamic Neural Networks and Genetic Algorithms. Masters thesis, Concordia University.

[img]
Preview
Text (application/pdf)
ShahrokhiTehrani_MSc_F2015.pdf - Accepted Version
7MB

Abstract

The main objective of this thesis is to propose and develop a fault detection, isolation and identification scheme based on dynamic neural networks (DNNs) and genetic algorithm (GA) for thrusters of the autonomous underwater vehicles (AUVs) which provide the force for performing the formation missions. In order to achieve the fault detection task, in this thesis two level of fault detection are proposed, I) Agent-level fault detection (ALFD) and II) Formation-level fault detection (FLFD). The proposed agent-level fault detection scheme includes a dynamic neural network which is trained
with absolute measurements and states of each thruster in the AUV. The genetic algorithm is used in order to train the DNN. The results from simulations indicate that although the ALFD scheme can detect the high severity faults, for low severity faults the accuracy is not satisfy our expectations. Therefore, a formation-level
fault detection scheme is developed. In the proposed formation-level fault detection scheme, a fault detection unit consist of two dynamic neural networks corresponding to its adjacent neighbors, is employed in each AUV to detect the fault in formation. Each DNN of the fault detection unit is trained with one relative and one absolute
measurements. Similar to ALFD scheme, these two DNNs are trained with GA. The simulation results and confusion matrix analysis indicate that our proposed FLFD can
detect both low severity and high severity faults with high level of accuracy compare to ALFD scheme.
In order to indicate the type and severity of the occurred fault the agent-level and formation-level fault isolation and identification schemes are developed and their performances are compared. In the proposed fault isolation and identification schemes, two neural networks are employed for isolating the type of the fault in the thruster of
the AUV and determining the severity of the occurred fault. In the fist step, a multi layer perceptron (MLP) neural network categorize the type of the fault into thruster blocking, flooded thruster and loss of effectiveness in rotor and in the next step a MLP neural network classify the severity into low, medium and high. The neural networks
in fault isolation and identification schemes are trained based on genetic algorithm with various data sets which are obtained through different faulty operating condition
of the AUV. The simulation results and the confusion matrix analysis indicate that the proposed formation-level fault isolation and identification schemes have a better performance comparing to agent-level schemes and they are capable of isolating and identifying the faults with high level of accuracy and precision.

Divisions:Concordia University > Gina Cody School of Engineering and Computer Science > Electrical and Computer Engineering
Item Type:Thesis (Masters)
Authors:Shahrokhi Tehrani, Shaghayegh
Institution:Concordia University
Degree Name:M.A. Sc.
Program:Electrical and Computer Engineering
Date:September 2015
Thesis Supervisor(s):Khorasani, Khashayar and Abdollahi, Farzaneh
ID Code:980426
Deposited By: SHAGHAYEGH SHAHROKHI TEHRANI
Deposited On:02 Nov 2015 17:07
Last Modified:18 Jan 2018 17:51
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

Research related to the current document (at the CORE website)
- Research related to the current document (at the CORE website)
Back to top Back to top