Research On Fault Prognosis Method For Key Components Of Helicopter Transmission Based On Deep Learning

As an important subsystem of helicopter,the health of helicopter transmission system directly affects the safety performance of helicopter.Due to long-term operation in high-temperature and heavy-load conditions,the key components of transmission systems such as bearings and gears are prone to failure.Therefore,it is of great significance to study the fault prognosis method for key components of the helicopter transmission system,which is beneficial to maintain equipment performance and avoid catastrophic accidents.However,most traditional data-driven prognostics methods are limited by the prior knowledge and expert experience,while the deep learning has the advantages of adaptive feature learning and strong data fitting,which provides a new idea for implementing the condition monitoring and fault prognostics of the helicopter transmission system.In this thesis,the fault prognosis method is studied based on deep learning for the rolling bearing,one of the key components of the helicopter transmission system.The main research contents include:(1)A summary of the common deep learning models applied in mechanical prognostics fault.The network structure,basic ideas and application of various deep learning models are introduced in detail,which lays a theoretical foundation for the following research work.(2)The degradation feature extraction method based on Convolutional Autoencoder(CAE)is studied.For the bearing run-to-failure data,the CAE is used to adaptively extract multiple degradation features.And these features are comprehensively evaluated to select the optimal features based on the three indicators of monotonicity,correlation and robustness.Finally,the effectiveness of the method is verified by comparing with the time domain statistical features and fusion features in the experiment and the noise-added simulation experiment.(3)The prognostics method based on Long Short-Term Memory(LSTM)network is studied.After obtaining the degradation feature trajectory,the short-term degradation state prediction and Remaining Useful Life(RUL)prediction are carried out based on the LSTM network.And the performance of the prognostics model is evaluated by the several evaluation indicators.In addition,The practicableness and effectiveness of the proposed method are verified by comparative experiments.In summary,this thesis mainly studies the degradation feature extraction,degradation state and RUL prediction method based on deep learning.The results show that on the premise of sufficient available data,compared with traditional fault prognosis methods,the fault prognosis method based on deep learning is efficient,universal and can obtain better prediction results.