A semi-supervised RUL prediction with likelihood-based pseudo labeling for suspension histories

Abstract

Accurate remaining useful life (RUL) prediction is an essential for efficient maintenance. In recent years, with the rapid development of industrial big data, many data-driven methods for RUL prediction have made significant progress, especially using deep learning. However, most of the proposed deep learning models only utilize labeled data and require a large amount of labeled data. In practice, the component of equipment is often replaced with a new one before it fails by preventive maintenance, resulting in a small number of failure histories and a large number of suspension histories. In other words, we have a small amount of labeled data and a large amount of unlabeled data. This paper proposes a new semi-supervised RUL prediction method using pseudo labels with flexibility in model architecture and low computational cost. For each suspension history, optimal pseudo labels are estimated using a likelihood-based method that takes into account important constraints, which enables more effective use of the information in both failure and suspension histories. The experiments on the C-MAPSS dataset validate the prediction accuracy of the proposed approach and provide several insights.