Research On Data-Driven Fault Diagnosis And Performance Degradation Prediction Of Rotating Machinery

As key basic parts in rotating machinery,bearings and gears are widely used in key fields such as ships,petrochemicals,electric power and metallurgy,and their performance,life and reliability play a decisive role in the safe and stable operation of the entire rotating machinery system.With the rapid development of signal detection and processing technology and computer technology,modern equipment monitoring is more and more inclined to multimeasurement points,high-frequency sampling,often collecting massive raw data with multimodality,low density,emergence and other characteristics.Machine learning-based fault diagnosis and degradation prediction models have a high dependence on high-quality label data,and training models using low-quality data often have problems such as underfitting and overfitting,resulting in a significant reduction in the accuracy of fault diagnosis and performance degradation prediction.Therefore,it is of great significance to carry out the research on early condition monitoring signal preprocessing,high robustness fault diagnosis model and performance degradation prediction model under non-ideal data of rotating machinery.This thesis uses machine learning as a technical means and takes key components of rotating machinery such as gears and bearings as research objects to carry out data-driven signal preprocessing,fault identification and performance degradation prediction research.From the perspective of the principle and application of machine learning,a combination of theoretical analysis,method research and experimental verification is used to construct a rotating machinery data mining,fault diagnosis and performance degradation prediction model,and verify its effectiveness.The main research contents of this thesis are as follows:(1)A preprocessing method for early condition monitoring data of rotating machinery is proposedIn order to overcome the dependence of machine learning-based diagnosis and degradation prediction models on a large number of high-quality label sample data.Firstly,an experimental scheme for collecting vibration signals for bearing and gear faults based on VALENIAN-PT500 vibration test platform was designed.Secondly,a clustering algorithm based on moving grid density(MGBDCN)is proposed to realize the pre-classification of fault data.Then,based on generative adversarial network(GAN)and transfer learning network,the fault dataset is enlarged.Finally,the effectiveness of the algorithm is verified on the fault signal of the test bench.(2)A fault diagnosis method of shrinking stack noise reduction auto-encoder based on embedded fusion is proposedSince the fault signal contains strong noise and accompanying vibration information.Therefore,an Embedded Fuse-Constriction Stacked Denoising Auto-Encoder(EF-CSDAE)with higher embedded fusion accuracy in noisy environment and feature recognition accuracy is proposed.Firstly,the failure forms of common rotating machinery are introduced;Secondly,the Jacobi penalty matrix is added to the common noise reduction self-coding loss function to make the network more robust;Finally,the original signal is embedded in the deep autoencoder network,and the shrinkage stack noise reduction auto-encoder structure of embedded fusion is constructed,and the algorithm finiteness is verified by experiments.(3)A performance degradation prediction method based on sliding window-long-term and short-term generative adversarial networks is proposedTo achieve the prediction of the remaining useful life(RUL)of rotating machinery under incomplete data conditions.Firstly,based on nuclear principal component analysis(KPCA)and local retention projection(LPP)algorithm,the performance degradation index system of rotating machinery is created.Secondly,the proposed Sliding Window-Long-Short term Generative Adversarial Network(SW-LSGAN)algorithm is used to expand the missing signal,and the RUL prediction of rotating machinery is realized,and verified on the NASA turbine engine degradation dataset.(4)Realized the development of a data-driven rotating machinery health management systemIn order to respond to possible failures during the real-time operation of rotating machinery and grasp the health status of equipment in real time,we have developed a datadriven rotating machinery health management system.Firstly,the overall architecture design of the system developed in this thesis is introduced.Secondly,the implementation and development process of the main functional modules of the system are elaborated in detail.Finally,the practicability of the development system was tested on the vibration test bench.