Research On Data Processing And Analysis Technology Of Equipment Maintenance Support

With the continuous development of modern manufacturing machines towards complexity,digitization and intelligence,rotating machinery is more widely used in industrial production.Bearings are the most used and most easily worn parts of mechanical equipment,and their health status has an important impact on the performance,stability and service life of mechanical equipment.Equipment maintenance support data processing and analysis methods represented by fault diagnosis,health assessment and remaining useful life prediction came into being.Through in-depth research in this field and in view of the current fault diagnosis and remaining life prediction algorithms require specific prior knowledge and a large amount of human intervention for feature extraction,and the Softmax classification capabilities of convolutional neural networks(CNNs)are not as good as emerging machine learning classification methods.This article mainly studies the following aspects:(1)A summary of equipment maintenance support data analysis and processing methods.Through the research of equipment fault diagnosis and remaining life prediction algorithm,the current development status of the algorithm is introduced,and the related algorithms are compared and discussed,and the problems to be solved are clarified.(2)Design of equipment fault diagnosis scheme.Aiming at the problem that CNN’s Softmax layer classification ability is not as good as the emerging machine learning classification method,a bearing fault diagnosis method based on continuous wavelet transform and Alex Net-Light Gradient Boosted Machine(Alex Net-LGBM)is proposed.The method can be divided into three parts:(1)Vibration signal data processing based on continuous wavelet transform: use continuous wavelet transform to extract time-frequency features from the original vibration signal of the bearing and convert it into a two-dimensional image of 32×32 pixels.(2)For fault feature extraction,this paper improves the Alex Net model to extract fault features from the time-frequency spectrogram,and compares it with the feature extraction capabilities of Le Net-5 and Efficient Net-B0 models.(3)For fault diagnosis,the extracted fault features are classified by Light Gradient Boosted Machine(Light GBM/LGBM)classification algorithm,and Bayesian optimization is used to select the optimal model parameters.A comparative experiment was carried out on the bearing data set of Case Western Reserve University(CWRU).Compare the various combination methods of Alex Net,Le Net-5 with multi-Grained Cascade Forest,LGBM,and Cat Boost.The results show that the Alex Net-LGBM fault diagnosis method based on continuous wavelet transform proposed in this paper has the best fault diagnosis accuracy.(3)The design of the prediction scheme for the remaining useful life of the equipment.A prediction method of bearing remaining useful life based on continuous wavelet transform and Alex Net-Gaussian Process Regression(Alex Net-GPR)is proposed.First,continuous wavelet transform is performed on the vibration data to generate a 90×90 pixel time-frequency spectrogram.Improve Alex Net into a regression model suitable for life prediction to predict bearing health index.The estimated value of the health index is eliminated and smoothed,and then the health index prediction curve is fitted through Gaussian process regression to predict the remaining service life of the bearing.Compared with other research results on the PRONOSTIA bearing degradation data set,the superiority of the proposed method is verified.