Civil Aviation Engine Fault Diagnosis Method Based on Deep Learning

As the core of aircraft, civil aviation engines have complex operating conditions and high reliability requirements, which pose severe challenges to fault diagnosis technology. Traditional diagnostic methods are often limited by problems such as missing data, feature complexity and diagnostic accuracy, and are difficult to meet actual needs. To this end, this paper introduces an intelligent diagnosis method combined with deep learning, taking Quick Access Recorder (QAR) data as the research object to comprehensively improve the efficiency and accuracy of fault identification. In data preprocessing, a multivariate interpolation algorithm is used to compensate for the incompleteness of QAR data, and wavelet transform and normalization techniques are used to optimize data features. In the modeling stage, a diagnostic model based on a fully connected neural network (FNN) is designed and trained, and the robustness and generalization ability of the model are improved through hyperparameter optimization. Finally, the model performance is verified and evaluated with the help of the Cross-Validation (CV) method. The experimental results show that by introducing the weighted cross entropy loss function, the data imbalance problem is effectively alleviated and the model's ability to identify minority class fault data is improved. At the same time, the use of the Adam optimizer accelerates the convergence process, and the training loss drops rapidly from the initial 0.540 to 0.240 in the 20th round.