Research On Adaptive Fault Features Extraction And Intelligent Diagnosis Of Rotating Machinery

Rotating machinery plays a decisive role in national production and life.Effective monitoring and diagnosis of the rotating machinery can ensure the safe and stable operation of the equipment finally avoid major accidents,which has great significance.Traditional intelligent fault diagnosis methods to extract features from a large number of data often rely on personal experience,which lacks adaptability and fails to make full use of the effective information of the original signal.Therefore,to realize fault detection and identification of key components of rotating machinery,this paper carries out research on the adaptive fault feature extraction method and intelligent diagnosis method based on machine learning.The research of this paper is as follows:(1)Aiming at the problems that machine learning algorithms often require a large amount of label data for supervised learning and the efficiency and time-consuming of iterative learning,an intelligent diagnosis method based on extreme learning machine auto-encode(ELM-AE)unsupervised feature learning is proposed.Firstly,ELM-AE is used to perform unsupervised learning on the collected label-free vibration data;Then use the network weights in the trained ELM-AE as the convolution kernel,and use the convolution pooling structure of the one-dimensional convolutional neural network to vibration data samples are used for feature extraction.Finally,Support Vector Machine(SVM)is used as a classifier to achieve fault classification and identification.The bearing/gear data set collected by the bearing and gear fault simulation test bench was used to analyze the proposed method.Results showed that the proposed method can perform adaptive feature extraction from the time domain data,which while ensuring higher fault recognition accuracy and effectively reduced network training time.(2)The fully connected neural network is difficult to effectively extract the local impact feature information in the vibration signal.The different vibration response frequencies caused by different mechanical faults,resulting in the poor diagnosis results of using the entire frequency band signal for feature extraction directly.To overcome these problems an intelligent diagnosis method based on back-propagation neural networks(BPNN)for learning multi-scale features was proposed.Firstly,the wavelet multi-scale transformation was used to obtain sub-signals of different frequency scales;Then,the label data information was used to extract multi-scale features from these sub-signals using BPNN;Finally,the SVM was used to realize fault classification and recognition of rotating machinery.The effectiveness of the method was analyzed using the rolling bearing data of the bearing and gear fault simulation test bed and the rolling bearing data of Case Western Reserve University in the United States.The results show that the multi-scale features of the different fault samples extracted by the method were distinguishable and can be used for effective identification of rolling bearing health status under different speed conditions.(3)Aiming at the problem that the information obtained by a single sensor is limited and it is difficult to fully characterize the operating state of the equipment,an intelligent diagnosis method based on Hilbert-Full vector spectrum and stack auto-encoders(SAE)is proposed.Firstly,the Hilbert-Full vector spectrum technique was used to fuse the vibration information collected by two vertically placed sensors,and use the Hilbert-Full vector spectrum main vibration vector as the input of the SAE to perform adaptive fault feature learning;Then,a small amount of label data samples are used to fine-tune the network model;Finally,the Softmax classifier is used to realize the fault classification and recognition of homologous dual-channel information.The proposed method was analyzed using SQI motor fault simulation experiment data set,and compared with the fault diagnosis method based on single-channel sensor signal,which verified the effectiveness and superiority of the method.