Remaining life prognostics of rolling bearing based on relative features and multivariable support vector machine

Life prognostics are an important way to reduce production loss, save maintenance cost and avoid fatal machine breakdowns. Predicting the remaining life of rolling bearing with small samples is a challenge due to lack of enough condition monitoring data. This study proposes a novel prognostics model based on relative features and multivariable support vector machine to meet the challenge. Support vector machine is an effective prediction method for the small samples. However, it only focuses on the univariate time series prognosis and fails to predict the remaining life directly. So multivariable support vector machine is constructed for the life prognostics with many relative features, which are closely linked to the remaining life. Unlike the univariate support vector machine, multivariable support vector machine considers the influences among various variables and excavates the potential information of small samples as much as possible. Besides, relative root mean square with ineffectiveness of the individual difference is used to assess the bearing performance degradation and divided the stages of the whole bearing life. The simulation and run-to-failure experiments are carried out to validate the novel prognostics model. And the results demonstrate that multivariable support vector machine utilizes many kinds of useful information for the precise prediction with practical values.