Fault Diagnosis Of Rolling Process Based On Neural Network

With the rapid development of automation technology in steel production,the rolling process is prone to major safety accidents.In order to ensure that the production process can run continuously and stably,it is particularly important to warn of faults and identify the type of fault.Due to the large amount of data generated by the strip process operation,faults are dependent on subjective experience when they occur and cannot be handled precisely.A neural network-based fault diagnosis method for the rolling process has been proposed.proposed to complete the diagnosis of complex equipment.The information of 21 variables such as rolling force and bending roll force in the finishing part of hot rolling has been extracted.For the data with large noise signal,wavelet packet is used to process the data.For the large amount of data and many kinds of variables,the data are processed using principal component analysis to remove the principal components with a contribution rate lower than 92% and reduce the 21-dimensional variables to 6 dimensions.The results show that the noise signal is reduced and the rolling feature principal components retain 95% of the information,which can improve the training speed of the rolling process network.For the problems of low efficiency and slow processing speed of rolling process fault diagnosis,a PCA-LSTM-based rolling process diagnosis model is proposed.The PCA-LSTM diagnostic model is trained using rolling feature principal components,and the diagnostic model is constructed with the fault category as the output.The results show that the PCA-LSTM diagnostic model can reduce the number of iterations,fast network training speed,diagnostic error lower than 1%,and good diagnostic effect.For the complex hot rolling process of strip steel,where the fault points are not easy to be quickly identified,a method of tracing the rolling process based on linear regression of neural networks is proposed.The optimal network parameters are found using the least squares method,and the abnormal points in the rolling data are identified by comparing the relative errors.The results show that the method can achieve rolling process traceability with a correct rate of 99% and have process practicality.Figure 30;Table 13;Reference 52...