Fault Diagnosis Research For Rotating Machinery Based On Vibration Signal Analysis

The reliability and stability of machinery operation affect the efficiency of manufacturing process.Therefore,the research of mechanical equipment fault diagnosis will maintain the safe operation of machinery and ensure the final product quality.The working condition of key parts seriously affects the change of mechanical equipment performance.It is always the research focus of mechanical equipment fault diagnosis to classify and evaluate the severity of the key parts of the mechanical equipments.Based on the theory of signal processing and artificial intelligence,this thesis takes the key parts of the mechanical equipment as the object and carries out research on several key problems of fault diagnosis research for the rotating machinery based on vibration signal analysis.While,fast processing and uncertainty identification of high-dimensional features in single-sensor information analysis is studied in depth.On this basis,multi-sensor information analysis based on deep learning and compressed sensing theory have been investigated in the implementation of in-site and remote mechanical fault diagnosis.The main research work and innovation results in this thesis are as follows:(1)In order to extract the key characteristics of the mechanical vibration signal in the background noise environment and realize the dimension reduction of the high dimension feature information,a supervised learning method based on the compressed decomposition features is presented.On one hand,the variational mode decomposition is used to solve the basis function selection,mode aliasing and edge effect,and background noise elimination.On the other hand,principal component analysis is applied to process Shannon spectrum entropy features from the decomposed signal,reducing the dimensionality and computation cost.It realizes the pattern recognition and finally improves the fast processing ability of the high dimension feature information.(2)In order to solve the problem of parameter uncertainty and model uncertainty in sensor signal processing,feature extraction,and learning process,an identification method of operation state uncertainty based on generalized interval is proposed.First,the variable mode decomposition method based on the generalized interval is used to process the acceleration sensing signal to obtain the frequency domain signal related to the fault characteristics.According to multiscale permutation entropy calculation,a generalized hidden Markov model is applied to identify the fault types and fault severity.The quantitative analysis of the aleatory and epistemic uncertainties is realized,and the robustness of the fault diagnosis is improved.(3)In order to overcome the dependence on priori and professional knowledge in the multi-sensor data feature extraction and fusion process,and solve the shortcomings in feature learning using traditional artificial intelligence method with shallow structure,a state identification method based on multi sensing information fusion and feature deep-learning is proposed.First,multiple acceleration sensors are used to collect the state information of a single stage gearbox.Then,the feature-level fusion of statistical features and Shannon energy spectrum features is carried out.Furthermore,it is processed by the deep learning method based on the stacked auto-encoders.Finally,the different crack depth of the gear is identified and classified effectively.(4)In order to solve the constraints of the traditional Shannon sampling law on sensing information processing and data transmission capacity of the condition monitoring system,and avoid raw data distortion caused by data compression and decompression,a compressed sensing-based condition monitoring and fault diagnosis method is carried out.First,the original mechanical vibration signals are compressed directly by the measurement matrix,so as to simplify the process of signal processing and feature processing.On the basis of this,on the one hand,the structured classification matrix is used to classify the compressed data.On the other hand,the compressive sampling matched pursuit method is used to restore the compressed data.Combined with the fault diagnosis method based on single and multi sensing information analysis,the comprehensive evaluation of mechanical equipment performance is carried out.This thesis will provide reference and reference for the practical application of the compressed sensing theory in mechanical fault diagnosis.