Kalman filter based fault detection and diagnosis

Author: Wei Cui

Cui, Wei, 2018 Kalman filter based fault detection and diagnosis, Flinders University, College of Science and Engineering

Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.

Abstract

In many control systems, sensors and actuators are integral parts, which can be utilized to detect the change of the system and take the needful action to achieve the control purpose. However, sensors and actuators have their own service life and can be damaged by various factors, which results in financial loss and casualties. These phenomena, when sensors and actuators stop working, are called sensors and actuators faults.

In this thesis, a Kalman Filter based Fault Detection and Diagnosis (FDD) scheme is proposed to detect and isolate different actuator faults for a given three-input three-output plate structure resonant system. This system can be used in many areas, such as doing operation on an emergency vehicle. The mathematical model of the given resonant system is first modelled and obtained by using open loop transfer function method and then changed into a discrete-time State Space Representation (SSR) model for further design purpose.

By setting the initial estimated state and its corresponding estimation error covariance, the following estimated state can be calculated by using the real-time control signals and measured outputs using the technique of Kalman filter. To utilize the Kalman filter technique for FDD purpose in a resonant system, the estimated outputs can be calculated using the estimated state. Next, a set of corresponding output error residuals can be generated by comparing the difference between the estimated outputs and measured outputs. A normalization algorithm calculating the RMS value of the corresponding residuals is applied to determine a threshold value to identify the location of actuator faults occurred in the system.

The concept of Kalman filter technique is first tested in an artificial system. Simulation results indicate that the estimated states generated by the Kalman filter can quickly approach the actual state values and track the actual state values all the time. The given 3*3 plate structure system with one fault is tested via simulation in MATLAB SIMULINK, which validates the proposed Kalman filter based FDD design. In the FDD design, four identical Kalman filters are constructed to form a Kalman filter bank for no fault, as well as actuator 1, 2 and 3 fault cases respectively. The estimated output of each case is computed by using the Kalman filter bank estimated state. The residuals, which are produced using estimated outputs and measured outputs, are analysed for the single fault detection and diagnosis. Then multiple actuator fault cases are introduced to the plate structure system and tested via MATLAB SIMULINK. The simulation results show that the corresponding multiple faults are detected successfully by analysing the output residuals using the proposed normalization algorithm.

After the proposed Kalman filter based FDD scheme is validated in simulation, it is then tested in a real-time experiment. Two discrete-time system models, a Kalman filter bank and a normalization algorithm are built to construct the experimental Kalman filter FDD scheme. One of the discrete-time system models is to produce the discrete-time measured outputs without fault, which is used for fault detection. The other is introduced to compute the measured outputs (possible with fault), which is used for fault diagnosis. The residuals computed in the Kalman filter bank for both single actuator fault cases and multiple actuator faults cases are generated. A set of residual data is recorded, and its RMS value is compared with the set threshold constant. The result verifies that all the actuator fault cases are isolated successfully.

Keywords: Kalman filter, vibration control, fault detection, fault diagnosis

Subject: Engineering thesis

Thesis type: Masters
Completed: 2018
School: College of Science and Engineering
Supervisor: Associate Professor Fangpo He