Prognostics and Predictive Maintenance Optimization Based on Combination BP-RBF-GRNN Neural Network Model and Proportional Hazard Model

Abstract

Owning to the advantage of keeping the operating environment safe, high reliability, and low production cost, predictive maintenance has been widely used in industry and academia. Predictive maintenance based on degeneration state mainly studies the degeneration prediction. However, on account of the error of the sensor and human, condition monitoring data may not directly reflect the true degeneration. The degeneration model with dynamic explanatory covariates which is named as proportional hazard model is proposed to deal with the semi-observed monitoring condition. And the degeneration prediction mainly adopts a single prediction model, which leads to low prediction accuracy. A combination forecasting model can effectively solve the above problem. Compared to the traditional prediction method, the neural network model can use the “black box” characteristic to indirectly construct the degeneration model without complex mathematical derivation. Therefore, we propose a combination BP-RBF-GRNN neural network model which is applied to improve the degeneration prediction with dynamic covariate. Based on the above two aspects, a predictive maintenance optimization framework based on the proportional hazard model and BP-RBF-GRNN neural network model is proposed to improve maintenance efficiency and reduce maintenance costs. The simulation results of thrust ball bearing show that the proposed method can effectively improve the degeneration prediction accuracy and reduce the maintenance cost rate to a certain extent.