Joint Prediction of Remaining Useful Life and Failure Type of Train Wheelsets: Multitask Learning Approach

Abstract

Train wheel failures account for disruptions of train operations and even a large portion of train derailments. Remaining useful life (RUL) of a wheelset measures how soon the next failure will arrive, and the failure type reveals how severe the failure will be. RUL prediction is a regression task, whereas failure type is a classification task. In this paper, the authors propose a multitask learning approach to jointly accomplish these two tasks by using a common input space to achieve more desirable results. A convex optimization formulation is developed to integrate least-squares loss and negative maximum likelihood of logistic regression as well as model the joint sparsity as the L2/L1 norm of the model parameters to couple feature selection across tasks. The experiment results show that the multitask learning method outperforms both the single-task learning method and Random Forest.