Author: Bo Sun
Sun, Bo, 2018 Fault Detection and Diagnosis for Vibration Isolation System Using Parity Relation Technique, 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.
In modern industries, most systems are constructed by multiple components with different functionalities. Only when all components function as they are designed to be, the system can perform effectively. Once a fault exists in the system, it may cause productivity deterioration or even system failure. The goal of this project is to fulfil fault detection and diagnosis (FDD) for additive faults on a vibration isolation system. In this project, FDD processes based on the parity relation technique for a multiple-input multiple-output (MIMO) vibration control of a mechanical structure is designed and examined.
A top plate bonded with three feet on a base plate, which forms a three-input three-output system, is used in this project to demonstrate the design of the FDD process. The control goal is to keep the top plate vibration free while the base plate is shaken consistently by a disturbance signal. By introducing different fault cases into the system, the control performance of the previously designed positive position feedback may not provide the best control results. Therefore, identifying when and where the fault occurs becomes essential for the fault accommodation.
A system identification study is conducted to experimentally obtain the mathematical model of the plant (i.e., the top plate together with the sensors and actuators attached to the plate). The transfer function matrix of the plant is then acquired through curve fitting techniques. Based on the design criteria of parity relation, the transfer function matrix is used to produce the state-space representation (SSR) of the plant that is further converted into the discrete-time form.
FDD process is a significant feature in fault tolerant control architecture. The FDD processes designed in this thesis are based on the parity relation technique, which is a popular mode-based fault detection technique that uses the residual as a fault indicator. When a discrete system is fault free, the designed residual of the system should be stable and close to zero in value. However, when a fault is introduced into the system, the residual will alter to indicate that there exists error(s) in the system. The core part of the parity relation technique is to design a parity vector which can eliminate the effects of the unknown state X(). The corresponding residuals can reflect whether the system works in a fault free condition or not. To detect fault, firstly one will need to define an order s of the parity space, which affects the performance of the FDD process. Secondly, the parity vector in this parity space will be designed based on the condition: ∗ = 0, where is a deterministic matrix associated with the state (). The designed residual () is calculated based on the measurable system input () and output (), and is not affected by the state ().
Using the parity relation technique, FDD processes respectively for a SISO system case and a MIMO system case are designed and validated through MATLAB Simulink. Firstly, a random system is chosen and is injected with a pre-defined single fault. Simulation studies are carried out to show that the parity residual value corresponding to the fault indeed changes from zero to non-zero, which validates the design principle of the parity relation technique. Secondly, the designed parity relation method is implemented to the plant model. Simulation results show that the used parity relation technique is able to theoretically detect pre-defined faults for the plant.
Upon the validation of the used parity space technique, an actuator-and-sensor fault isolation is implemented through designing different parity vectors. A set of structured residuals are generated where each residual is sensitive to all faults apart from one corresponding fault. The corresponding fault will then be located through a designed fault-code table. To carry out the fault isolation examinations, studies on isolating individual sensor faults among the three sensors of the plant are first conducted, and the corresponding scenarios are implemented in MATLAB Simulink. Simulation results show that the set of designed residuals has an excellent performance in detecting single sensor faults in the real plate system. A similar procedure is arranged for the isolation of individual actuator faults among the three actuators of the plant. However, simulation results reveal that the effect of the actuator fault isolation is not as clear as that of the sensor fault isolation. Due to the existing physical properties of the given plant, the parity vectors designed respectively for Actuator 2 fault and Actuator 3 fault of the given plant appear to be very close to each other. This leads to the difficulties in clearly distinguishing the responses of these two residuals. It is anticipated that further modification of the designed parity vectors that can take into account the existing physical properties of the given plant may produce a more reliable FDD process applicable to all types of individual actuator faults effectively.
In conclusion, the studies carried out in this thesis have proven that in principle, the parity relation technique can be implemented on a MIMO vibration cancellation system successfully for additive faults’ detection and diagnosis. Comparing with other methods, the implementation of the FDD processes using the parity relation technique is simple and flexible. Using the proposed FDD processes, the sensor or actuator fault occurrences of the given plant can be effectively detected. Furthermore, the individual fault among the three sensors of the given plant can be diagnosed successively. Although the proposed method can only provide partial individual fault diagnoses among the three actuators of the system due to the physical properties of the given plant, the general concept of the design is verified via a specific actuator fault scenario. Future work includes the modification of the proposed FDD process design method to enable it to diagnose all individual actuator faults successfully, and to be physically implemented to the real laboratory experimental rig of the given plant.
Keywords: Fault Detection, Fault Diagnosis, FDD, Vibration Isolation System, Parity Relation
Subject: Engineering thesis
Thesis type: Masters
Completed: 2018
School: College of Science and Engineering
Supervisor: Fangpo He