Research On Remaining Useful Life Prediction And Fault Diagnosis Of Mechanical Equipment Based On Deep Learning

Mechanical equipment is the main component of a ship,and its smooth running is of great significance to the safety and reliability of ship operations.However,mechanica l equipment is affected by factors such as complex working environment,improper operating conditions,etc.,it is prone to structural damage,health degradation and other problems,which can cause major safety accidents and huge losses.In order to ensure the safety and reliability of ship operations,it is urgent to carry out research on the remaining useful life(RUL)prediction and fault diagnosis technology of mechanical equipment.With the rapid development of artificial intelligence technology,especially deep learning,exploring the application of deep learning to remaining useful life prediction and fault diagnosis of mechanical equipment is a topic that is very worthy of attention.This thesis focuses on the key technology of RUL prediction and fault diagnosis of mechanical equipment based on deep learning.The main research contents and innovations are as follows:(1)To improve the generalization performance and prediction accuracy of the current RUL prediction methods,an ensemble learning framework that integrates deep bidirectio na l recurrent neural networks(DBRNNs)is proposed.In this framework,a number of DBRNNs with different structures are constructed to realize the long-term correlation of monitor ing time-series data from multiple dimensions in the front and back directions,and a new customized loss function is designed to evaluate the training bias of the network model;then,an ensemble learning algorithm based on multiple regression decision trees is used to fuse the prediction results of the network model to improve the overall prediction accuracy and generalization performance of the model.Finally,an engine dataset is used to verify the proposed method.The results show that the proposed method can accurately predict the RUL of the in-service engine,and its performance is better than existing methods.(2)In view of the problem that the current fault diagnosis methods require a large amount of labeled fault data and are only effective for specific operating conditions,a mechanical equipment fault diagnosis method based on deep transfer learning is proposed.In this method,the adversarial training mechanism is introduced to extract high-dimensio na l feature expression from the source domain and the target domain,and the adaptive layer is used to make the feature distribution of the source domain and the target domain closer.In addition,the proposed method can also make full use of limited data to accurately diagnose faults for mechanical equipment under different working conditions and different working scenarios.The method is verified on rolling bearing and gearbox.The results show that the proposed method can extract domain invariant features from vibration signals directly,and achieve high-precision fault diagnosis of mechanical equipment under a variety of operating conditions and a small number of unlabeled fault samples.(3)Aiming at the problem that the existing diagnosis model is difficult to be applied to the unknown faults of equipment,a self-growth training method of fault diagnosis model based on category incremental learning is designed.This method adopts a specia l "knowledge distillation" technology to design a cross-distillation loss function,and adaptively update and adjust the existing fault diagnosis model.While ensuring that the model’s diagnostic accuracy for existing categories does not decrease,self-learning of new fault categories is realized,and the self-growth of fault diagnosis model knowledge is achieved.A set of gearbox multi-fault simulation experiments is carried out to verify the proposed self-growth model.The results show that The proposed method can realize the self-growth learning of fault diagnosis model,and the model can still maintain high fault diagnosis accuracy while greatly reducing the training time.