Estimation Research In Remaining Useful Life Prediction Driven By Degradation Data

Equipment failure caused by sudden accidents often lead to disasters. To ensure the safety of the service of machinery and equipment served in nuclear power, aerospace, military and other fields is of great significance, accident-before prediction can significantly reduce the accident rate to avoid casualties.While timely and accurate life prediction for machinery and equipment is often difficult because of the system complexity and nonlinearity. It is difficult to measure the actual parameters affected by external factors and stochastic uncertainty. In addition, the system parameters and status also showed statistical characteristics in stochastic process.Therefore, trying to find out the internal parameters and system status for the remaining useful life prediction based on external observation data with the influence of stochastic uncertainty, estimation theory need to be researched.The estimation theory are studied Based on different system degradation processes, Include the following:First, remaining life prediction methods are classified, and the Rul prediction methods driven by statistical data are focused to explore. Different estimation methods are formulated and compared, and the relationship between these estimation methods are analyzed.Second, modeling, estimation and life prediction based on linear Wiener degradation process is researched.Modeling through state-space model and parameter estimation methods for the Wiener degradation process with random drift effect is accomplished. Kalman filter and maximum expected-Kalman filtering for parameter estimation is compared, and the bearing vibration signal data is used for estimation and life prediction verification.Third, modeling, estimation and life prediction based on nonlinear Wiener degradation process is researched.Build state space model through nonlinear wiener process, further suppose parameters obey different distribution and research for hyper-parameter estimation based on particle filter is accomplished, and the battery capacity degradation data is used for estimation and life prediction verification.Finally, the relationship between remaining useful life prediction and parameter/state estimation in different degradation processes is summarized.The work in the paper might offer reference for the modeling and estimation of remaining useful life predictions.