Prediction Of System Residual Life Based On Kernel Density Estimation

With the manufacturing industry gradually shifting from "manufacturing-centered" to "service-centered",the safe and reliable operation of system equipment has become particularly important.The sudden failure in the normal operation of system equipment may lead to the stop of the whole production process and cause significant economic losses,or even cause danger to people.If the remaining life of the system can be predicted before the failure of the equipment,it can be timely and actively preventive maintenance measures,so as to avoid the occurrence of accidents.Therefore,it is of great significance to predict the remaining life of the system equipment and then formulate a reasonable and effective maintenance plan.At present,large-scale system equipment is developing towards automation,large scale and integration,which leads to very complex and changeable faults of system equipment,expensive manufacturing cost of large-scale system equipment,and it is difficult to carry out a large number of life tests.Therefore,the prediction method based on physical model is difficult to be applied to system equipment.However,a large amount of operation data and state degradation data are generated during the operation of system equipment.Therefore,a data-driven prediction method can be used to predict the remaining life of the system equipment.In this paper,the vibration data obtained from the fatigue life test of the gear box is used to represent the degradation state of the gear box,and the residual life of the gear box is predicted based on the fault threshold based on the kernel density estimation method.The main work and research results are as follows:(1)A kernel density residual life prediction method with adaptive window width is proposed.First,in view of the nuclear estimate fixed window width would lead to inaccurate prediction problem,based on the local density of sample improved adaptive selection window width,by K-nearest neighbor method to calculate the distance between the sample and then judge the density of the local samples,namely in the area of different density to select differentwindow width to improve nuclear estimate the accuracy of the prediction results.Secondly,every time a new sample data is added to the traditional kernel estimation model,the kernel density estimation based on these samples has to be recalculated,which causes the problem of repeated calculation and large amount of calculation.In view of this problem,a model of real-time updating of kernel density estimation is established,and the real-time updating of degradation distribution and prediction of residual life distribution is realized after new sample degradation data are obtained through real-time monitoring.Finally,the effectiveness of the proposed method is verified by gear degradation test.(2)A remaining life prediction method considering the degradation state of system equipment is proposed.Due to the degradation process of the system in the health state,the degradation data are different when the system is in different health states.The degradation state and degradation mode often change continuously with time and sample data.In order to realize accurate prediction in the process of system degradation,a method is proposed to classify the sample data and then divide the degradation state mode to realize adaptive residual life prediction considering the equipment degradation mode.