UMP Institutional Repository

Hybrid fault detection using kalman filter and neural network for quadrotor micro aerial vehicle

Chan, Shi Jing (2018) Hybrid fault detection using kalman filter and neural network for quadrotor micro aerial vehicle. Masters thesis, Universiti Malaysia Pahang.

[img]
Preview
Pdf
Hybrid fault detection using kalman filter and neural network for quadrotor micro aerial vehicle - Table of contents.pdf - Accepted Version

Download (454kB) | Preview
[img]
Preview
Pdf
Hybrid fault detection using kalman filter and neural network for quadrotor micro aerial vehicle - Abstract.pdf - Accepted Version

Download (122kB) | Preview
[img]
Preview
Pdf
Hybrid fault detection using kalman filter and neural network for quadrotor micro aerial vehicle - References.pdf - Accepted Version

Download (192kB) | Preview

Abstract

This thesis introduces the application of time-domain Hybrid Fault Detection (HFD) methods for application in a quadrotor Micro Aerial Vehicle (MAV). The application aims to solve one of the main problems of the quadrotor, which is its inability to reach the exact target location that the user intended. The problem may be due to a faulty signal happened in the sensor or the actuator side or both, causing the quadrotor unable to complete the task given. Among the reasons of the faulty signal are the occurrences of signal in quadrotor, in the sensor or actuator side, as well as possible communication problem. When actuator fault, sensor fault or both faults occur, the controllers cannot function well and hence its performance reduce. At the initial control design stage of quadrotor, it is usually designed under the assumption that no faults would occur in quadrotors. The Faulty Detection (FD) method is therefore crucial to ensure quadrotor system can work properly and efficiently. The proposed method for the fault detection in this study uses hybrid technique which combines the extended kalman filter and artificial neural network (ANN). Two classes of approaches are analysed: the fault system identification approach ANN and the observer-based approach using the extended kalman filter. The extended kalman filter recognizes data from the sensors of the system and indicates the residuals of the system in the sensor reading. Residuals prediction is based on the fault magnitude and the time occurrence of fault. The information will then be fed to ANN, which consists of a bank of parameter estimation that generates the failure state. ANN is an algorithm that is used to determine the fault condition and determine its severity in the quadrotor system. ANN is designed based on nonlinear autoregressive network with exogenous inputs (NARX) scheme so that it can be trained to generate output based on the simulation behaviours of the quadrotor. The different fault locations are used as input vectors for training an artificial neural network (ANN). The result of the residual signal before filtration and after filtration showed that Kalman-ANN is able to identify single fault as well as multiple faults. For all individual faults including the multiple fault detection, the accuracy of the detection is 78.89 percent. It can be conclude that the newly proposed hybrid FD method in this thesis is able to accurately detect the location fault, for both the sensor and actuator faults simultaneous in the quadrotor.

Item Type: Thesis (Masters)
Additional Information: Thesis (Master of Science) -- Universiti Malaysia Pahang – 2018, SV: DR. DWI PEBRIANTI, NO. CD: 11581
Uncontrolled Keywords: Quadrotor; Kalman filter; Artificial Neural Network; Fault Detection; Fault Isolation; Kalman-ANN; Kalman-Fuzzy; Nonlinear Autoregressive Network with Exogenous Inputs
Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering
Faculty/Division: Faculty of Electrical & Electronic Engineering
Depositing User: Mrs. Sufarini Mohd Sudin
Date Deposited: 29 May 2019 07:19
Last Modified: 29 May 2019 07:19
URI: http://umpir.ump.edu.my/id/eprint/24600
Download Statistic: View Download Statistics

Actions (login required)

View Item View Item