Marine Diesel Engine Fault Diagnosis By Improved Multiple Model

The stable operation of marine diesel engines is the guarantee for the safety of ship navigation,but the harm caused by the failure of diesel engines due to complex systems and various mechanisms cannot be ignored.For the existing problems in fault diagnosis of marine diesel engines,this thesis combines modelbased and data-based fault diagnosis methods,proposes a fault diagnosis method based on improved multi-model.The main work is as follows:Firstly,in terms of the difficulty of entity analysis of marine diesel engines,a mechanism model is established to replace the entity and realize fault simulation based on thermodynamics,fluid dynamics,etc.Analyze the operating characteristics of each part of the diesel engine and choose the volume method or the average value model in the zero-dimensional to model,combined with the bench test data and the parameter identification method to improve the expression ability of the average value model.The simulation results show that the maximum error of the model is 6.687% compared with the experimental data,which could meet the requirements of operation simulation and fault simulation.Secondly,regarding the problems of slow diagnosis speed and poor robustness of the multi-model method based on Kalman filtering,an improved multi-model fault diagnosis method based on probability fusion is proposed.The posterior probability of the support vector machine and the conditional probability of the multi-model are fused based on Bayesian linear melding.The joint diagnosis of the two methods is realized,and the improved method is verified by a fault simulation case.The results of fault simulation shows that the improved multi-model after probabilistic fusion improves fault detection and isolation speed by 14.17% and 3.79%,respectively.Thirdly,according to the high dimension and large amount of calculation of the state model of the marine diesel engine,this thesis establishes a directed graph of the entire engine based on graph theory combined with the mechanism model,and the directed graph is decomposed based on the observability.The corresponding state model group is established in the sub-graph.Then combined with the state model group and the improved multi-model method,a framework of marine diesel engine fault diagnosis is established.The running time of each subsystem is checked,the result shows that the maximum time required for filtering in the subsystem is 1.166 ms,which is 69.86% less than that of the whole diesel engine model.Fourthly,use the mechanism model to simulate fault which diagnosed by multi-model and improved multi-model methods respectively.Inspect the improvement effect of the method based on probabilistic fusion by single fault,and the weight value of the probability fusion is discussed.Inspect the effectiveness of the fault diagnosis framework by multiple faults,and the influence of the fault coupling between the subsystems after decoupling is analyzed.The experimental results show that the improved multi-model improves the diagnostic speed and robustness,and the fault diagnosis framework can effectively isolate the impact of parameter fluctuations caused by faults between different systems.This thesis establishes a fault diagnosis framework based on improved multiple-model and graph theory decoupling,which is of great significance to ensure the stable operation of marine diesel engines.