Research On Intermittent Fault Diagnosis And Prognosis Of Hybrid Systems Based On Hybrid Bond Graph And Intelligent Optimization Algorithms

In modern industrial productions,many industrial systems can be treated as hybrid systems with the interaction of continuous dynamics and discrete dynamics.With the increasing demand for system safety and reliability,hybrid system health monitoring(including fault diagnosis and prognosis)has become an important research field.Intermittent faults are noncontinuous faults and can recover required functions without repairments.In hybrid systems,intermittent faults will occur not only in continuous components but also in discrete components.Discrete components only include two states(i.e.,ON and OFF)without corresponding physical parameters.The intermittent fault degradation of discrete component is usually reflected in duration or frequency.However,there is no quantification method of the degradation process of intermittent faults of discrete components,resulting in the case that the discrete component prognosis under intermittent faults cannot be accomplished.The intermittent fault degradation process of continuous component can be reflected in duration and magnitude simultaneously,which are both required to be considered for the prognosis of continuous components.Moreover,the intermittent fault possesses stochasticity in appearance,disappearance,and magnitude.At present,intermittent fault diagnosis and prognosis have attracted extensive attention of researchers,while the relevant researches of intermittent fault diagnosis and prognosis for hybrid systems are very limited.Therefore,based on hybrid bond graph,intelligent optimization algorithms,and extreme learning machine,this dissertation studies the intermittent fault diagnosis and prognosis method of hybrid systems.The main research contents and contributions of this dissertation are as follows:(1)The hybrid bond graph based modelling method of hybrid systems and the theoretical basis of hybrid system fault diagnosis using diagnostic hybrid bond graph model are introduced.Firstly,the hybrid bond graph model of hybrid system is established.According to the sequential causality assignment procedure for hybrid system diagnosis,the diagnostic hybrid bond graph model can be built from the hybrid bond graph model.Then,independent global analytical redundancy relations and independent augmented global analytical redundancy relations can be derived from the diagnostic hybrid bond graph model for fault detection of discrete components and continuous components.After that,the mode change signature matrix is constructed for discrete fault isolation,and the fault signature matrix is constructed for continuous component fault isolation.Finally,the feasibility of the proposed methods is verified using a simply hybrid circuit system.(2)The intermittent fault diagnosis and prognosis method of discrete components in hybrid circuit system is studied.Firstly,the diagnostic hybrid bond graph model of hybrid circuit system is established,from which the independent global analytical redundancy relations can be derived for discrete component fault detection.Secondly,combining independent global analytical redundancy relations through the algebraic operation,the dependent global analytical redundancy relations are obtained.Under the multiple discrete faults condition,with independent and dependent global analytical redundancy relations,the integrated mode change signature matrix based discrete fault isolation method is developed.Then,the Levy flight Biogeography-based Optimization based discrete component intermittent fault estimator is designed to identify fault appearing and disappearing instants of discrete components under intermittent faults.After that,with the obtained fault estimation results,the quantitative method of discrete component intermittent fault features and degradation process based on tumbling window concept and coordinate reconstruction approach is proposed.The discrete component prognosis under intermittent fault is realized based on the Weibull degradation model and based on the degradation model selection method,respectively.Finally,the effectiveness of the proposed methods is verified on the hybrid circuit system experimental platform.(3)The intermittent fault diagnosis and prognosis method of continuous components in electric scooter system is studied.Firstly,the diagnostic hybrid bond graph model of electric scooter system is built,from which independent global analytical redundancy relations and independent augmented global analytical redundancy relations are derived for fault detection of the monitored electric scooter system.Secondly,the mode change signature matrix is constructed,based on which the discrete fault identification method is proposed for checking whether the electric scooter is in the faulty mode or not.When the system is in the normal mode,the extended fault signature matrix is established for continuous component fault isolation.Then,the adaptive competitive swarm optimization algorithm is developed for intermittent fault estimation of continuous components,where the fault estimation is to identify intermittent fault magnitude,appearing and disappearing instants of continuous components.After that,with the aid of tumbling window,the dual degradation processes of intermittently faulty components can be established,where the dual degradation processes include the duration degradation process and the magnitude degradation process.Consider the possible discrepancy between duration degradation rate and magnitude degradation rate,the dual degradation processes based joint remaining useful life prediction method is proposed for continuous components under intermittent faults.Finally,the effectiveness of the proposed methods is verified on the electric scooter system experimental platform.(4)The distributed intermittent fault estimation method of hybrid systems is studied.Concerning the disadvantages of the global system model based centralized intermittent fault estimation method(such as many unknown parameters to be identified,complex fitness function,heavy computational burden,and so on),the distributed intermittent fault estimation method is proposed using the structural model decomposition.The submodels are decomposed from the global diagnostic hybrid bond graph model through the structural model decomposition approach.According to the obtained submodels and intelligent optimization algorithms,the submodel based fault estimator can be constructed,by which the distributed intermittent fault estimation can be implemented.(5)Aiming at the situation that the degradation model of the intermittently faulty component is usually unknown,the improved extreme learning machine based prognosis method under intermittent fault is developed.Firstly,according to the fault estimation results,the intermittent fault feature data set can be established with the aid of tumbling window.Then,the obtained intermittent fault feature data set is used to train the extreme learning machine prediction model,where the adaptive competitive swarm optimization is used to optimize input weights and hidden layer biases of standard extreme learning machine to obtain the optimal extreme learning machine.Finally,the proposed method is verified based on the electric scooter system,where the intermittent fault magnitude degradation process of the continuous component is considered in the experiments.