A data-driven fault diagnosis fusion algorithm and health management framework was proposed to solve the problems of asynchronism, feature redundancy and instability of weak fault recognition in multi-source monitoring data of aero-engine. Method The method takes 21 types of sensor variables and three types of working condition parameters as input. After abnormal correction, normalization, and sliding window coding, the channel sensitivity prior, adaptive weight allocation, and error feedback correction mechanism are introduced to form a fusion feature for fault classification, diagnostic confidence calculation, and health state assessment. The experimental results show that the model Accuracy is 96.37%, F1-score is 95.74%, AUC is 98.21%, and the average inference time is 19.8 ms. Compared with Transformer, the F1-score is improved by 2.55 percentage points and the inference time is reduced by 8.1 ms. Ablation results show that the adaptive weight and error correction contribute significantly to the performance of the model, which can support the stable transformation of fault identification results to maintenance strategy output.
