Multiple-model-based Convolutional Neural Network for Machine Remaining Useful Life Prediction under Multiple Operation Modes

Due to the changes in external environments and inner settings, industrial machines are always switched between multiple operation modes, causing multimodal-distributed records for monitoring variables. The data with diverse distributions will cause difficulties for machines’ remaining useful life (RUL) prediction tasks. To address this problem, a new multiple-model-based prediction method is proposed in this article. During the data preprocessing, a new data compensation strategy is designed to eliminate mode-specific biases of monitoring data, which is used to alleviate the influence of multimodal-distributed data caused by multiple operation modes. Besides, a multiple-model-based CNN model is designed to focus on the feature extraction of monitoring data with different degradation degrees with different sub-feature extractors. During online monitoring, proper features are selected from these sub-features to calculate the final RUL of the monitored machine. The proposed method is verified on the C-MAPSS dataset, and the experiment results show the effectiveness of the proposed method.