Fault Diagnosis Of Multi-Temperature Measuring Device For Hot Strip Steel Production Line

In the hot strip rolling production process,the strip temperature has an important influence on the strip quality and is the most important parameter to determine the processing properties and physical properties of the finished strip.The strip steel production line is mainly distributed with five temperature measuring devices:after the heating furnace,after the rough rolling,before the finishing rolling,after the finishing rolling and before the coiling.The environment in which the temperature measuring device is located is relatively harsh,making it one of the most troublesome parts of the rolling production system.When the temperature measuring device fails,the actual temperature of the strip will not be obtained,so that the quality of the rolled product will not reach the standard.When the temperature measuring device fault,the real temperature of the strip steel will not be obtained,so that the quality of the rolled product will not reach the standard.Therefore,it is necessary to diagnose the fault of the multi-temperature measuring device on the hot strip steel production line.In this paper,the temperature after rough rolling is taken as the starting point,and the temperature before coiling is the end point.The fault diagnosis mechanism of multi-temperature measuring device combining single-process fault diagnosis and production line is proposed.The BP neural network and deep confidence network method are used to establish the fault diagnosis models of the three temperature measuring devices before,after and after the rolling,and the initial diagnosis of the operating state of the three temperature measuring devices is completed in the paper.First of all,In order to realize the accurate prediction of the strip temperature of the temperature measurement points in each process,the temperature prediction model of the temperature measurement points of each process was established,and use the difference between the predicted value and the measured value of the model as the fault residual to perform fault diagnosis.When two methods are used to establish the temperature prediction model,it involves the determination of the input and output parameters of the model,the normalization of the data,the determination of the model structure,and the evaluation of the network.Among them,the most important is the training and testing of the model structure(the BP neural network is the number of hidden layers and hidden nodes,the DBN model is the number of RBM and the number of hidden nodes),and the optimal model structure is obtained.Secondly,the two methods are compared and analyzed.The comparison between BP neural network and deep confidence network can be used for the prediction of strip temperature,and the latter is excellent in forecast accuracy and forecast stability.Finally,the fault diagnosis of the two fault diagnosis methods is carried out to verify the feasibility of the two methods for the fault diagnosis of the temperature measuring device.The rolling temperature of each process has a relationship before and after.The strip temperature measurement value of the previous process will be used as the input of the fault diagnosis model of the latter process.Therefore,the final state of the four temperature measuring devices needs to be judged in conjunction with the relationship between the temperature of each strip in the fault diagnosis model.In this paper,the fault diagnosis logic of the multi-temperature measuring device of the production line is proposed.When there is a continuous fault,the predicted value of the rolling temperature of the previous process is used instead of the measured value as the input method of the next process fault diagnosis model for fault diagnosis.In order to compare the diagnostic effects of the two methods,the fault diagnosis is carried out by taking the fault diagnosis of the temperature measurement device before the finishing rolling and the coiling.The results show that although the two methods have little difference in fault diagnosis accuracy,the DBN-based method is superior to BP neural network in fault diagnosis stability.Combined with the former,the accuracy and stability of temperature prediction are better than the latter.Therefore,the DBN-based method is chosen to establish a fault diagnosis mechanism for multi-temperature measurement devices combining single-process fault diagnosis and production line.Finally,the fault simulation of multi-temperature measuring device is carried out.It is known from the simulation results that the fault diagnosis scheme proposed in this paper can be used for fault diagnosis of multi-temperature measuring device in hot strip steel production line.