Research on one-dimensional data quality assurance for machinery health monitoring using compressive sensing and enhanced context encoder

Reconstruction of abnormal data to improve data quality is of great importance for machinery health monitoring (MHM). Existing data reconstruction methods are generally limited by strict assumptions, such as signal sparsity and random sampling, along with high computational costs, making them poorly adaptable to MHM data. To assure high-quality MHM data, this study developed a novel enhanced context encoder based on compressive sensing (CS-ECE). The loss function utilized in CS-ECE is designed to consider multiple feature dimensions, enabling accurate reconstruction of signal characteristics in both time and frequency domains, which can significantly enhance the reliability of MHM results. The proposed CS-ECE’s effectiveness and superiority are confirmed through real-world data collected from a high-speed train and ablation studies. Comparative analysis shows that the proposed CS-ECE yields a higher fitting degree and lower error levels compared to four classical and five representative state-of-the-art CS algorithms, especially in terms of time cost, which is at least two orders of magnitude faster.