Remaining useful life estimation based on the hybrid Support Vector Regression with the Nonlinear Autoregressive with Exogenous Input (SVR-NARX)

Abstract

The rotating machines took an important role in the industries and manufacturing technology, the continually using of these tools leads to its breakdown, which manages to several loess, including high economies loss. This paper aims to avoid the unexpected failure of those tools by estimating the Remaining Useful Life (RUL) of the ball bearing, for this sake a couple of methods namely Decision Tree (DT), and the hybrid Support Vector Regression (SVR) with the Nonlinear Autoregressive with Exogenous Input (NARX) named as SVR-NARX which is applied to determine the RUL, first Time Domain Features (TDF) are extracted from the raw vibration signal and then this TDF are selected using the DT method, after that the Discrete Wavelet Transform (DWT) is applied on the selected features to separate the high and low frequencies from the selected features, the extracted frequencies components (EFC) are used as input which are used to train and test the SVR-NARX, the obtained model is then used to determine the RUL, The online PRONOSTIA database is applied for the training and testing the SVR-NARX, the SVR-NARX is compared to its primitives the SVR and NARX trained and tested using the EFC and the original selected feature, the overall of the applied strategy indicate that the SVR-NARX trained by the EFC gave high results in terms of Root Mean Squared Error (RMSE=0.0090, 0.0085) and Factor of determination (R2 = 0.999, 0.997) for both training and testing respectively, the applied strategy gave high result which should be further considered for other machine related tasks.