A Structure Metadata Driven Framework for Agile Heterogeneous Sensor Integration in Industrial PHM System

In complex industrial equipment, the diversity of sensor types and the significant disparities in sampling rates pose challenges for multi-source data fusion. This study proposes a structural-meta-information-based method for agile access and unified modeling of heterogeneous sensors. The method achieves lossless data integration through high-frequency alignment and a masking mechanism, while incorporating time-frequency positional encoding and meta-information embedding to enhance structural and semantic representation across channels. In addition, a channel adaptive mechanism enables new sensors to be integrated without historical data and allows direct channel expansion without retraining. Validation on experimental data from a hybrid transmission system demonstrates that the proposed method outperforms conventional fusion approaches in terms of diagnostic accuracy, robustness to missing channels, and model scalability, indicating strong potential for engineering applications.