The Residual Life Prediction Model Based On Time Series Analysis

With the concepts of exact repair and re-manufacturing more deeply, predictingthe remaining life of the system or component become increasingly important. But nowmore and more reliable systems and components, it is difficult to obtain its failure data,in fact, regardless of the system or component, its performance decreases with use, sowe turn to consider the remaining life of degraded systems or components. This articlepredict the remaining life of the degradation system using time series method.First, the progress of the remaining life is reviewed, and the details of thedata-driven residual life prediction model are introduced, there is a method based ondirect observation data, including random coefficient model method, the residual lifeprediction method based on Wiener process, and method based on Gamma predictionprocess; random filtering method based on indirect observation data.Followed by the test data of a rolling bearing vibration intensity, we use AICcriterion to determine the number of auto-regressive moving average model (ARMA),and the parameters of the model estimated by using the least squares method to obtainthe predicted value of the remaining life, experiments have shown a greater predictionerror. Instead established state space equation, Kalman filtering algorithms to predict theremaining life of the bearing, experiments show that the predicted value and the actualmeasured values match well.Then the component degradation is monotonous, so the process of the Gamma-based state space model is established, under have no the initial information of thecomponents, using EM algorithm to estimate model parameters, obtain the probabilitydensity function of remaining life.Finally, summarizes this paper and indicate the need for further research solve theproblem.