| As an indispensable key component in industrial machinery and equipment,rolling bearings are related to the safety and stability of the equipment.As an important technology of rolling bearing prediction and health management(PHM),fault diagnosis and remaining life prediction can effectively identify the location of rolling bearing faults and estimate their remaining life,so as to realize early intervention and maintenance of mechanical equipment and ensure the safety of enterprises production.For bearing vibration signals with non-stationary and non-linear characteristics,continuous wavelet transform(CWT)is a time-frequency analysis method that can extract the time and frequency domain information of the signal and display it in time-frequency images;deep learning methods are widely used in image feature extraction and modeling analysis.Aiming at the feature extraction of rolling bearing vibration signals and the construction of a deep identification model,this paper proposes a deep learning method for rolling bearing health status assessment and remaining life prediction based on multi-dimensional features and time-frequency analysis.The specific research contents are as follows:(1)Research on rolling bearing fault diagnosis method based on multi-dimensional features.When modeling and analyzing rolling bearing vibration signals using shallow machine learning methods,the selection of input features is very important for the analysis results.For the problem of effective feature extraction of rolling bearing vibration signal,a fault diagnosis method of multi-dimensional features rolling bearing based on multi-domain parameters and transformation features is proposed.The signal time-domain and frequency-domain statistical parameters and time-frequency domain feature extraction methods EMD,WPD and complex envelope analysis are used to extract multi-dimensional feature information,which is combined with SVM classifier for bearing fault diagnosis.The diagnosis results are compared with the automatic extraction of the original bearing vibration signal features by auto-encoder(AE)into the SVM classifier.The experimental results show that the recognition rate of the fault diagnosis model with multi-dimensional features combined with SVM is higher than that of the AE-SVM method,with the highest fault recognition rate of 78.20% under single operating conditions and 70.26% under alternating operating conditions.(2)Research on rolling bearing fault diagnosis method based on time-frequency image and deep learning.Aiming at the problem of low recognition rate of bearing fault diagnosis due to inadequate extraction of vibration signal features in strong noise background of rolling bearings,a rolling bearing fault diagnosis method based on time-frequency images and deep learning is proposed.In this method,the original bearing vibration signal is converted into time-frequency image by CWT,the time and frequency domain information of the vibration signal is extracted,and the internal parameter-optimized Res Net model is used for bearing fault diagnosis.The experimental results show that the Res Net model can fully extract the bearing operating state information in the time-frequency image,and the fault diagnosis recognition rate of the model is 99.24%,99.90%,98.47% and 99.92% under four loads of 0hp,1hp,2hp and 3hp after 100 iterations of training under a single working condition;the fault diagnosis recognition rate of the model under alternating working condition also reaches90.56%.Compared with the shallow machine learning algorithm and the shallow neural network algorithm,the Res Net model has obvious advantages in the bearing fault diagnosis.(3)Research on deep learning prediction method for remaining life of rolling bearings.For rolling bearings in the process of remaining life prediction requires a large number and complex extraction of bearing life degradation features,and there are problems of insufficient feature extraction resulting in low accuracy of bearing life prediction.In this paper,we propose a deep learning-based method for remaining life prediction of rolling bearings,using CNN to extract deep features of time-frequency images containing bearing degradation information,and using Dropout layer to solve the problems of too many parameters and time-consuming information mining in CNN model training.The experimental results show that,compared with the LSTM model,the CNN model has higher accuracy and lower prediction error for bearing life prediction under different working conditions. |
