| Prognostics and Health Management includes four steps: data collection,health indicators construction,health stage division,and remaining useful life(RUL)prediction.Based on the acquired raw data of rolling element bearings,starting from the construction of health indicators,this paper divides the health stages of rolling element bearings,finds the fault occurrence time(FOT),and predicts its RUL.The main research contents of this paper are as follows:First,construct health indicators based on the characteristics of the original data.The original data of rolling bearing is a set of high-dimensional data,which is directly used as the input of the model,which will cause the calculation to be too complicated and cause the problem of "dimension disaster".In response to this problem,the Peak-to-Peak feature in the time domain is selected as the degraded feature,and when the Peak-to-Peak feature has high frequency noise and spikes,it is cumulatively enhanced to construct a health indicator.Secondly,the number of health states is determined according to the degradation trend of rolling element bearings,and the Viterbi algorithm is used to divide the health stages to clarify the fault occurrence time.In view of the characteristics of rolling element bearing degradation trend,the Odd-Even Half-Sampling criterion is used to determine the number of states,and the AIC criterion is used for two-way verification.On this basis,with the help of the Viterbi algorithm in the Hidden Semi-Markov Model(HSMM),the health stage is divided to determine FOT.Carry out envelope spectrum analysis on the vibration signal at the FOT,study the laws behind it,and determine its fault type.Then,the deep learning method of the Bidirectional Long and Short-Term Memory(BiLSTM)is used to improve the accuracy of the RUL prediction.Traditional neural networks have poor interpretability and fixed input sequence length,which will affect the accuracy of prediction.In order to solve these two defects,this paper introduces an Attention Model,and uses the percentage of the remaining life as the label to predict the remaining life.Finally,the corresponding evaluation indicators are given,and the method proposed in this paper is compared and analyzed with common methods through simulation experiments.The results show that the method proposed in this paper is more effective. |
