Remaining Useful Life Prediction For Hydraulic Pump Based On Hidden Semi-Markov Model

With the development of science and technology,reliability research has become a hot area.The reliability of military equipment is directly related to national defense.Hydraulic components are very important in military equipment.It is of great significance to research the remaining useful life of hydraulic components for improving the reliability of the entire system.This paper is dedicated to predicting the remaining useful life of hydraulic pumps.The main research includes:(1)The failure mechanism of the hydraulic pump was studied.The main failure mode of the hydraulic pump is the wear failure caused by particles in the contaminated oil.The influencing factors of the wear life of the hydraulic pump were analyzed,and the proportion of these influencing factors was given.In the experiment,the actual working conditions of the hydraulic pump can be simulated by adjusting the influencing factors.The wear test for the hydraulic pump was designed,and the basic test flow was proposed,which provide test data for the research of the remaining useful life of the hydraulic pump.(2)This paper showed the existing theory of residual life prediction based on hidden semi-Markov model.This paper proposed a residual life prediction model based on observation sequence prediction and state recognition.Firstly,a combined model of genetic algorithm and neural network was used to predict the observation sequence.The observation sequence at future time was used to identify the nearby state turning point,which makes each observation point has its unique remaining useful life.In addition,an improved sequential importance re-sampling method was proposed.This method can realize the state recognition function combing the model parameters of the hidden semi-Markov mode,and the result proves that this method has high recognition accuracy.(3)A combined model of genetic algorithm and neural network was used to predict the observation sequence.The global search ability of genetic algorithms can determine the optimal parameters of the neural network model.In this paper,the genetic algorithm was used to determine the learning efficiency and momentum coefficient of the neural network model to exert the maximum performance of the neural network.(4)An improved sequential important re-sampling algorithm for state recognition was proposed.Sequential importance re-sampling algorithm is based on the Markov process,but Markov process does not apply to Hidden semi-Markov models.Therefore,the state resident probability was introduced into the important probability distribution of sequential importance re-sampling algorithm,which solved the problem.(5)The prediction of the observation,the identification of the health state and the prediction of the remaining life were carried out,taking the hydraulic gear pump as the research object.The results show that the combined model of genetic algorithm and neural network has good performance for the prediction of the observation sequence,and the state recognition accuracy based on the improved sequential importance resampling algorithm is higher than that based on the hidden semi-Markov model library,and the remaining life prediction model based on observation sequence prediction and state recognition can realize the prediction function of the remaining life and have good prediction performance.