Research On The Edge-Cloud Collaborative Diagnosis Method And System Of Rolling Element Bearing

As one of the most common parts of rotating machinery systems,rolling element bearings are widely used in many industrial fields such as transportation,energy engineering,aviation,intelligent manufacturing,etc.They are the ‘industrial joints’ of rotating machinery.Therefore,it is of great significance to develop fault diagnosis methods and systems for rolling element bearings to ensure safe and stable equipment operation.The rotating speed of industrial equipment is often variable according to the working condition and requirement,which poses a challenge for the vibration-based condition monitoring and fault diagnosis.Another existing challenge is the class imbalanced problem.The data samples of normal state are usually abundant during machinery operation,while faulty samples are scarce.Facing the abovementioned challenges,this thesis proposed several methods to address the varying speed and class imbalance problem,and designed an edge-cloud collaborative diagnosis system for rolling element bearing.The details are as follows:1.In order to effectively diagnose faults of rolling element bearings under varying speed,an instantaneous phase estimation and order tracking method based on speed information is firstly proposed in this thesis.In this method,the instantaneous frequency and phase of rotating machinery can be calculated automatically by using speed information which is easy to obtain but not precise.Thus,the order tracking methods are easier to implement.Simulation analysis and experimental results show that the proposed method is able to estimate the instantaneous phase accurately,and provides precise order tracking results even if the given speed information has a large deviation.2.An online filtering procedure and diagnosis method based on Balanced Envelope Spectrum(BES)is proposed to detected fault of rolling bearings at speed varying condition.The principles of filter parameter for vibration signals are first discussed,and the requirements of optimization algorithm for online filtering is analyzed in this thesis.An online Protrugram procedure is designed to extract equivalent characteristic frequencies automatically.Based on the characteristics of Protrugram,the balanced envelope spectrum method is designed to realize automatic fault analysis of rolling element bearing.3.Aiming at the class imbalance problem caused by the scarcity of fault data,a learning framework named Deep Focused Parallel Convolutional Neural Network(DFPCN)is proposed.The reasons for the performance degradation of deep learning-based diagnosis methods are first discussed.A neural network model constructed by parallel convolutional architecture,optimized by the Adaptive Cross Entropy loss function(ACE Loss)is designed to perform the imbalanced classification.Experimental results on both constant and variable speed bearing dataset show that the DFPCN model greatly improves the diagnosis accuracy of minority samples,while ensuring the accuracy of the majorities.4.In order to understand the Convolutional Neural Networks(CNN)for fault diagnostics,a visualization method is proposed.By utilizing the Grad-CAM algorithm,the interested signal segments of CNNs are able to be extracted.The result answers which signal components are used as fault criteria in CNN models from the time domain.Besides,a kernel visualization method based on gradient ascent technique(GAK-vis)is proposed.GAK-vis is designed to projects the pattern of the deep convolutional kernels onto the input signal.It shows the processing pattern of deep kernels which usually difficult to obtain in the frequency domain.5.Based on the edge-cloud architecture,an edge-end hardware platform for signal acquisition and analysis is built,and an edge-cloud collaborative diagnosis system for rolling element bearings is developed in this thesis.The system integrates basic functions such as real-time data acquisition,online signal analysis and diagnosis,maintenance history management,cloud storage,etc.Thus,it can not only meet real-time analysis requirements,but also achieve collaborative maintenance at lower cost.