Research On Bearing Fault Diagnosis Of Annealing Kiln Roller Table Transmission System Based On Deep Learning

The annealing kiln roller drive system is in between the tin bath system and the cold end processing system,which is responsible for the function of glass transfer,during the transfer process,the glass is gradually cooled from high temperature to room temperature,and the high temperature working environment requires the annealing kiln roller system to have strong stability to ensure the continuity of production and the quality of glass.The spherical ball bearing is the key transmission component of the annealing kiln roller conveyor system,and its operating condition directly affects the smooth operation of the annealing kiln roller conveyor system,once the spherical ball bearing fails,the glass quality is not guaranteed.The paper focuses on the key technologies of bearing vibration signal noise reduction,feature extraction and pattern recognition,and builds a deep learning-based roller conveyor bearing fault diagnosis model and shaft bearing fault transfer diagnosis model,which can accurately discern the operating state of the bearings in the roller conveyor system of the annealing kiln.The main research contents of the paper are as follows:(1)To address the problem that the effective signal of annealing kiln roller drive system bearing is interfered by the strong mechanical background noise drowning,we propose the combination of Ensemble Empirical Mode Decomposition(EEMD)and Continuous Wavelet Transform(CWT)The signal noise reduction and time-frequency feature representation method.The method combines the correlation coefficient and the mean cliff criterion to filter the EEMD decomposition of the Intrinsic Mode Function(IMF),and then the filtered IMF components are superimposed and reconstructed with the residual components to complete the noise reduction,and then the CWT is used to transform the one-dimensional noise reduction signal into a two-dimensional time-frequency map,and the histogram equalization is used to enhance the time-frequency features.The validation results of rolling bearing test data and annealing kiln roller conveyor bearing data show that the data processing method proposed in the paper can effectively filter out the noise in the bearing signal of annealing kiln roller conveyor drive system and realize the time-frequency representation of the bearing monitoring signal.(2)For how to extract and accurately identify the effective features and fault states in the time-frequency diagram of roller conveyor bearings,an Inception-Lstm based roller conveyor bearing fault diagnosis model is proposed.The diagnostic model uses the Inception module to extract multi-scale fault features,and connects the Lstm network to further mine and learn the temporal dependencies existing among the features to enhance the feature extraction capability of the model.The verification results of rolling bearing test data and annealing kiln roller bearing data show that the constructed bearing fault diagnosis model has excellent feature extraction capability and recognition performance,and can accurately discern the operation status of roller bearings.(3)To address the problem that the shaft bearing fault samples are not easy to obtain,a roller conveyor bearing fault migration diagnosis model based on Multi-kernel maximum mean discrepancies(MK-MMD)is proposed.The diagnostic model uses MK-MMD as a metric function to reduce the difference in data distribution between the source and target domains,and then uses a parameter fine-tuning method to determine the optimal number of parameter fixed layers to optimize the whole migration diagnostic model.The validation results of rolling bearing test data and annealing kiln roller bearing data show that the constructed bearing fault migration diagnostic model can accurately discern the operating status of shaft bearings under the conditions of small label samples and cross working conditions.