Online Predictive Health Status Assessment On Industrial Equipment

Over time,mechanical equipment or its components are prone to failure.The optimization of mechanical device working time and the minimization of life cycle cost require on-line monitoring of its degradation process and reliable prediction of remaining life.In order to fulfill such critical needs,prognostics health management is a new engineering discipline,linking failure mechanism and life cycle management.The purpose is to extend the service life of mechanical equipment and reduce the cost of development and maintenance.With the development of sensor technology,a large number of sensors have been used in industrial equipment.Therefore,data driven health state management technology has been favored by more and more researchers.Data driven prognostics health management uses information from past,current and future of mechanical equipment to evaluate degradation state,diagnose faults,predict faults and calculate remaining life.Its main structure includes four key processes,namely,degradation feature selection,health state assessment,fault prediction and residual use life estimation.Based on these processes,this paper proposes a set of predictive health status on-line evaluation methods for industrial equipment.(1)In view of the massive high dimensional characteristics of degradation data,and at the same time,the degraded feature set should show the trend of continuous increase or decrease,robustness to monitoring noise and operation environment,this paper proposes an automatic degradation feature selection method based on genetic algorithm and evaluation criteria(monotonicity,trend and robustness).(2)In view of the dynamic operation environment and uncertain stochastic factors in the industrial equipment,the same equipment may have different health states.A SG-FCM dynamic health state assess method based on SCM,FCM and genetic algorithm is proposed.The initial clustering center of FCM is provided by SCM,and the influence of initial parameter selection on FCM performance is reduced.The ability of global search algorithm to optimize FCM climbing search method reduces the probability of entering the local optimal solution.(3)In view of the problem that the univariate prediction model can not adequately describe the complex degradation process of equipment,a multi-variable multi-step fault prediction model is established by using LSTM,and the best prediction model is selected by using multi parameter optimization method.Based on this prediction model and SG-FCM,the LSG-FCM dynamic residual life calculation method is proposed,whichcan be used as a machine with different number of health states.The device automatically assigns the failure threshold,so as to dynamically provide the end point of the prediction model and improve the accuracy of residual use life evaluation.Finally,the effectiveness of the proposed method is proved by the data set of aero turbine engine.The experimental results show that the feature selection method proposed in this paper can effectively select the best feature subset for tracking equipment degradation process.Based on the extracted feature subset,SG-FCM algorithm can get better clustering effect than traditional FCM,and dynamic health state assessment under different conditions.The prediction accuracy of dynamic residual life calculated by LSG-FCM model can reach 73%,and the prediction error span is [-20,42] smaller than other approaches.