Research On Fault Classification Method Of Inker Roller Bearing Of Offset Press Based On Autoencoder

At present,the printing industry has both the attributes of processing and manufacturing and the cultural industry,and is an important part of GDP.In the Thirteenth Five-Year Plan,the printing industry will continue to develop in the direction of green,intelligent,digital,reliable and stable printing equipment is an important foundation for the development of the printing industry.Rolling bearings are the key basic components in printing presses,which are very important to guarantee the performance of printing presses.In order to achieve long-term stable operation of printing presses,many scholars have proposed fault diagnosis techniques for bearing components of printing presses,aiming to prevent printing machinery faults.The economic losses brought by this paper take the rolling bearing in the inking roller of the offset press as the research object,and carry out unsupervised learning on the characteristics of the vibration signal of the bearing.The neural network structure is built to identify the type of failure.The main work of the paper is as follows:(1)The theoretical principles and algorithmic processes of sparse self-encoders and stacked self-encoders are studied,and they are applied to the data of Xi'an Jiaotong University-Shengyang database and the bearing database of West Chu University.The feature extraction ability of the two self-encoders to the original vibration signal is analyzed through experiments,and the classification accuracy rate of the two types of data fault types is calculated.It is proved that deep networks can mine deeper features than shallow networks.(2)A fault feature recognition method based on the combination of self-encoder and maximum correlation entropy criterion is proposed,which realizes the fault classification and recognition of the original signal,and the accuracy rate can be maintained at about 98%.The bearing vibration signal is input into the built network structure without any processing,and unsupervised learning is performed to extract the characteristics of the data through the network's own functions and structure to realize the classification of different types of data.Through experiments,the influence of parameters such as the number of hidden nodes,the number of input layers,and the learning rate on the network feature extraction ability is analyzed.(3)The proposed method is applied to the identification of printing machine bearing fault types.By extracting the fault characteristic data of the inking roller bearing of the offset press,the proposed method is verified,the optimal fault classification network model is obtained,and the noise is added to the signal to verify the network normalization ability.It is suitable for the classification task of printing machine bearing faults with noise interference,and the classification accuracy of the network can be maintained at a high level.Related research provides an important reference for the information acquisition and identification technology of intelligent fault diagnosis of printing presses,and has certain engineering application value.