Real-Time Remaining Useful Life Prediction of Cutting Tools Using Sparse Augmented Lagrangian Analysis and Gaussian Process Regression

Remaining useful life (RUL) of cutting tools is concerned with cutting tool operational status prediction and damage prognosis. Most RUL prediction methods utilized different features collected from different sensors to predict the life of the tool. To increase the prediction accuracy, it is often necessary to mount a great deal of sensors on the machine in order to collect more types of signals, which can heavily increase the cost in industrial applications. To deal with this issue, this study, for the first time, proposed a new feature network dictionary, which can enlarge the number of candidate features under limited sensor conditions, and the developed dictionary can potentially contain as much useful information as possible. This process can replace the installation of more sensors and incorporate more information. Then, the sparse augmented Lagrangian (SAL) feature selection method is proposed to reduce the number of candidate features and select the most significant features. Finally, the selected features are input to the Gaussian Process Regression (GPR) model for the RUL estimation. Extensive experiments demonstrate that our proposed RUL estimation framework output performs traditional methods, especially for the cost savings for on-line RUL estimation.