Research on multi-source operation and maintenance data fusion and deep residual network modeling for aviation equipment fault prediction

This study focuses on the fault prediction of aviation equipment, and proposes a modeling method combining multi-source operation and maintenance data with improved depth residual network (ResNet). Aiming at the multi-source heterogeneous characteristics of aviation equipment operation and maintenance data, the feature extraction and vectorization of structured, semi-structured and unstructured data are realized through data preprocessing, and the unified feature dimension of cross-modal feature alignment mechanism is constructed. The dynamic fusion strategy based on channel attention mechanism is introduced to adaptively integrate multi-modal features by weighting. In the improved depth ResNet, a multi-scale residual unit is proposed, different convolution kernels are used to extract multi-scale fault features, and a SE module is added at the end of the residual module to realize channel attention weighting, which enhances the sensitivity and robustness of the model to weak abnormal signals. At the same time, a mixed loss function combining classification and regression is designed to jointly optimize the tasks of fault classification and residual service life prediction. The experimental results show that this method is superior to the existing mainstream methods in fault identification accuracy, residual life prediction accuracy and model robustness, which can effectively improve the fault prediction performance of aviation equipment and provide strong support for preventive maintenance of aviation equipment.