Uncertain Model Based Fault Diagnosis For A Ship Rudder Electro-hydraulic Servo System

Rudder electro-hydraulic servo systems (RESS) play an important role in the control of ships. Keeping a normal and steady RESS operation is crucial to sea trip safety. Due to the complexity and harsh working conditions, RESS is often influenced by some time-varying and poorly known factors (such as water forces acting on the rudder blade, friction force from rudder components and disturbances). These uncertainties along with the inherent nonlinearities and parametric uncertainties of RESS complicate the design of an effective fault diagnosis system. Thus the model based fault diagnosis technology is no longer suitable. This thesis studies the uncertain model based fault diagnosis strategy for a ship RESS. Common faults existing in the RESS can be detected, isolated and reconstructed. Such studies could be helpful for development an online automatic fault diagnosis scheme in an actual RESS, which is important for the safty of a RESS.In this thesis, an uncertain model based disturbance decoupling method is proposed for an RESS. The diagnosis method is able to eliminate the influences of inherent nonlinearities, parametric uncertainties and unknown disturbances. Unlike other methods for similar systems, the proposed method offers several advantages:a) It does not require estimations of viscous damping, friction and disturbance, which are often difficult to measure accurately, and b) no boundary assumptions are made on disturbance. Through reasonably allocating the parameter matrices of the RESS observer, the robust fault diagnosis can be realized and the fault sensitivity will not be sacrificed. Numerical and experimental tests verify that all the fault diagnosis results will not be influenced by the unknown disturbance. Six different faults which can’t be diagnosed by the disturbance estimation method can be diagnosed by the proposed method. Based on the robust fault diagnosis strategy, a batch of specific observers is designed and work synchronously and cooperatively to isolate different faults in the RESS. Commonly encountered faults in RESS and added sensors (which are used for fault diagnosis) are all considered. Then a fault and disturbance synchronous reconstruction strategy is proposed. Compared with the indirect fault diagnosis strategy which often needs to extract specific fault features, the proposed fault reconstruction strategy can realize direct fault diagnosis, which is more useful to detect early gradually changed faults. Experimental results show that all the faults can be isolated effectively and the reconstruction error is within ±5%. Considering the real-time and economy requirements, several measures are taken:On one hand the optimized observer design algorithms are proposed, which could satisfy the limiting conditions easily and obtain the solutions quickly; on the other hand a weighted adaptive threshold method is introduced to save storage space and improve computation efficiency. A unified fault diagnosis scheme is then formulated in the context of an industrial RESS. An online fault diagnosis software is also developed and only two normal pressure sensors are added. Experimental results show that all the faults can be diagnosed in 1 second, thus the uncertain model based scheme is efficient for online use.The thesis is outlined as follows:In chapter 1, the RESSs and the commonly encountered rudder faults are introduced. Then the model based fault diagnosis technologies and the fault diagnosis methods for electro-hydraulic servo system are reviewed. Based on the above introduction, a preliminary fault diagnosis strategy for the RESS is presented and the main contents of the research are stated at last.In chapter 2, the derivation of the state equations and commonly encountered fault types of RESS is presented, and the model uncertainties are analyzed. An uncertain model based fault diagnosis idea is established after some numerical and experimental investigations. The numerical and experimental test rig is introduced and different faults are simulated and analyzed.In chapter 3, an uncertain model based disturbance decoupling method is proposed for the RESS. A certain model based method is also offered for comparison. The derivation and the design steps of the two methods are given respectively. Based on experimental tests, a weighted adaptive threshold method is introduced to improve decision making. The normal state performances, robustness and fault detection capabilities of the two methods are tested and compared numerically and experimentally.In chapter 4, a batch of observers is designed and work synchronously and cooperatively to isolate different faults in the RESS. Commonly encountered faults in RESS and added sensors (which are used for fault diagnosis) are all considered. The design principles and steps of the isolation observers are presented. The optimized observer design algorithms are also proposed, which could satisfy the limiting conditions easily and get the solutions quickly. Then an online robust fault diagnosis scheme is formulated. Numerical and experimental tests validate the effectiveness of the fault isolation scheme at last.In chapter 5, based on the structure and the robustness of the designed observers, a fault and disturbance synchronous reconstruction strategy is presented. The reconstruction algorithms are derivated for different faults in the RESS. Numerical and experimental tests validate the effectiveness of the reconstruction scheme.In chapter 6, the research work and important results are summarized. The main contributions of this dissertation are highlighted. Future work is presented in the end.