A Remaining Useful Life Prediction Framework Integrating Multiple Time Window Convolutional Neural Networks

摘要

Efficiently predicting Remaining Useful Life (RUL) of equipment is fundamental for assessing system health and developing maintenance strategies. Considering the high degree of inconsistency among the length of degradation trajectories, a multiple time window Convolutional Neural Networks (CNN) based framework is introduced to improve prediction accuracy. In this proposed method, one-dimensional CNN is adopted to learn degradation trends from historical sensor data, and a multiple time window strategy is exploited to reduce training errors and increase the utilization rate of test data. The performance of this proposed framework is validated through an experimental study and compared with state-of-the-art models. The comparison and analysis have demonstrated that this framework can achieve the best overall performance, and thus it can provide strong support for preventive maintenance.