Degradation Trend Estimation of Rolling Bearing Based on Multivariable LSSVM over Time Slices

Degradation trend estimation of rolling bearing is widely applied in many engineering applications. With the rapid development of sensors, massive data will be acquired because of the monitoring of machinery. So how to extract and use the effective data from the big data for trend estimation has profound research value. The least square support vector machine (LSSVM), which is a kind of novel artificial intelligence technology, has scored tremendous achievements in this field. Among kinds of methods, multivariable LSSVM takes a full consideration of multiple variables to improve the accuracy of prediction. However, it may ignore the influence factors with time. To improve the performance of multivariable LSSVM, a kind of multivariable LSSVM over time slices is proposed which considers both the influence of multivariable and volatility of data. Firstly, several sensitive features are chosen to construct a new training sample which include the information of data volatility. Then LSSVM is employed to construct the model for prognosis. The vibration signals generated by an accelerated bearing run-to-failure experiment are employed to validate the effectiveness of the proposed method and the results show that the multivariable LSSVM over time slices outperforms general multivariable LSSVM.