Research On Fault Diagnosis Methond For Rolling Bearing Based On Dense Convolutional Network

As the core supporting component of rotating machinery,the actual working condition of rolling bearing is often complex,such as high speed,high temperature and high load.And it is also often affected by all kinds of shock and vibration,which leads to high failure rate.The traditional fault diagnosis method usually needs to disassemble the bearing from the equipment and carry out fault diagnosis combining with the actual experience of the engineer,which results in a long diagnosis cycle and poor timeliness,which can not meet the needs of real-time monitoring.At the same time,in the process of vibration signal acquisition,the disturbance of external environment noise and the imbalance of samples caused by the lack of fault samples make the fault diagnosis process of rolling bearing more difficult.In order to solve the problem that environmental noise may be introduced in the process of vibration signal acquisition,this paper propo ses an adaptive WTD method based on sample entropy,and denoises the signal data.This method decomposes the vibration signal by DWT technology.Then,based on the sample entropy of each layer of wavelet coefficients,different threshold functions are assigned to each layer of coefficients and the coefficients are filtered.Finally,the noise reduction of vibration signal is completed by inverse discrete wavelet transform of the processed wavelet coefficients,and the noise reduction method is verified by relevant experiments.In order to solve the problems in the process of bearing fault diagnosis,such as the poor timeliness of diagnosis and the unsatisfactory accuracy of diagnosis,this paper introduces the Dense Net model to extract the characteristic signals which can represent the running state of the bearing from the mixed vibration signals of the bearing.At the same time,with the help of EMD algorithm and PCA algorithm,the model is improved,and a rolling bearing fault diagnosis system based on EP-Dense Net model is constructed,and completed the corresponding fault diagnosis work.At the same time,through a series of comparative experiments,the diagnosis system is verified.In order to solve the problem of sample imbalance in fault diagnosis,this paper proposes a rolling bearing fault diagnosis method to improve the diagnostic accuracy of the system in case of sample imbalance.Based on the existing Dense Net model,combined with CS technology,the system sets different misclassification cost weights for each sample according to the number of samples of each fault category,and integrates the cost weights into the objective function of the integration framework to solve the problem of sample imbalance from the algorithm level,so as to improve the final diagnosis accuracy of the system.And through a series of comparative experiments,the diagnostic system was validation.Finally,combined with the actual needs of bearing fault diagnosis,the application-level implementation of the methods and models proposed in this paper and the development of the diagnostic system were carried out in Python,and the system was applied to the actual signal data to complete an operational demonstration,which verified the diagnostic system and provided the correspond ing theoretical basis and technical support for the enterprise’s bearing fault diagnosis work.