Study On Fault Diagnosis Methods Of Electro-hydraulic Position Servo System In A Vehicle Mounted Howitzer

Vehicle-mounted howitzer(VMH)is the leading weapon in the future local war.Accelerating the development of high mobility,more powerful and high firing accuracy vehicle-mounted artillery is the main direction in the future.Electro-hydraulic position servo system(EHPSS)is one of the core components of the VMH firepower system.Its function and performance directly affect the response speed,safety,and stability of the VMH.In the complex battlefield environment,the EHPSS may have varying degrees of fault,resulting in its operational function and performance degradation,and even security risks.To ensure the safe and stable operation of VMH,this article studies the relevant fault diagnosis methods for typical faults of the EHPSS.The methods not only include the design theory and design method based on unknown input observer,sliding mode observer,and support vector machine,but also provides theoretical and technical support for the solution of practical engineering faults.The work is as follows:(1)The working principle and basic structure of the EHPSS are analyzed systematically.Based on the failure mode and effect analysis,the variant failure modes,effects and causes of each subsystem are summarized.The EHPSS model and its fault model of the VMH are established.The validity of the simulation model is verified by comparing it with the experimental results.The main uncertain factors affecting fault diagnosis are found by simulation.(2)Considering the nonlinear characteristics,parameter uncertainty,and system noise of the EHPSS,a fault detection method based on unknown input observer is used.A nonlinear unknown input observer is designed to decouple the disturbance from the fault,and the residual of the observer is only sensitive to the fault.The dynamic threshold residual evaluation method based on statistics is used to reduce the probability of false alarm and missing alarm.The online detection of cylinder leakage fault,servo valve fault and power source pressure fault is realized on the experimental platform.The experimental results verify the effectiveness of the method.(3)In view of the existence of modeling error,parameter perturbation,and output noise disturbance,the sliding mode variable structure technology is applied to the fault detection and diagnosis of actuator and sensor.The actuator fault detection with output disturbance is considered,and the system is transformed into two subsystems.Sliding mode observers are designed respectively to decouple the actuator fault and output disturbance,and the actuator fault is reconstructed.A sliding mode observer group is designed to detect and separate the sensor faults.Finally,the sensor faults are reconstructed to obtain detailed information about the size and shape of the sensor faults.(4)For the case of the known fault category labels in the samples,a supervised learning fault diagnosis scheme based on standard support vector machine is proposed.Firstly,the training time and accuracy of different kernel functions are compared,and the kernel function of radial basis function is selected to construct support vector machine classifier.Then,grid search method,genetic algorithm,particle swarm optimization algorithm and simulated annealing algorithm are used to optimize the kernel function parameters and penalty coefficient respectively.The optimized parameters of particle swarm optimization algorithm are used for model training.The test data set with fault is input into the trained model,and whether the sample contains fault is determined according to the accuracy of prediction.Finally,the "one-to-one(OVO)" separation strategy is adopted to construct a classifier group composed of six classifiers.The slight fault data samples and severe fault data samples are input respectively.The task of off-line fault detection and separation under supervised learning is realized..(5)For the case of the lack of fault category labels in the samples,the fault detection method based on unsupervised learning of one-class support vector machine(SVM)is combined with adaptive affinity propagation and Gaussian mixture model to realize fault detection and separation.Firstly,the data samples that need to be detected are preprocessed.The parameters and training samples optimized by PSO are input into the one-class SVM model to obtain the training model.Then,The test data set containing faults is input into the trained model to predict the fault category.Subsequently,the four kinds of fault data samples are composed of sequences,and input into the adaptive affinity propagation to obtain the number of clusters.The Expectation Maximization(EM)iterative algorithm is used to solve the parameters of the Gaussian mixture model.The maximum weight parameter of the single Gaussian model component determined that this component is the current fault.Under the condition of unsupervised learning,multi-fault offline detection and separation targets are realized.