Surface defect prediction of hot rolled strip based on SAE-DBN hybrid depth network

The surface quality defects of hot rolled strip seriously restrict the high-end product quality. The surface quality defects of the strip have the characteristics of diversity and randomness,and the formation mechanism of different defects is different. The process that causes surface quality defects of strip steel is complex,and it is difficult to effectively control the surface quality defects. To solve the problem that it is difficult to realize online diagnosis of the quality defects of hot rolled strip,the diagnosis and prediction research is carried out for the defects of iron oxide scale indentation,edge warping and edge crack that occur frequently in the hot rolling production process. Based on the analysis of surface defect mechanism,the cause variables that affect the surface quality defects of strip during rolling process are defined,and they are used as the input data source of the prediction model of surface quality defects of hot rolled strip. Then,based on depth confidence network and depth sparse self-coder,a surface defect prediction model of hot rolled strip based on SAE-DBN hybrid depth network is proposed. Based on the DBN diagnostic model,the optimal network weight is obtained by training a single SAE. Assign the obtained network weight to the first layer RBM of the DBN model,and initialize the network weight and bias in this way,to further improve the prediction ability and learning efficiency of the model and improve the robustness of the depth self-coder. The model was validated using actual data from the hot strip production process,and the results showed that the average prediction accuracy of the SAE-DBN hybrid depth network defect prediction model reached 94.23%. Finally,the edge warping defect of hot rolled strip is taken as the contrast object of BP neural network diagnosis model,DBN diagnosis model,and DSAE diagnosis model. Compared with the three models,the prediction accuracy has been improved by 18.56%,12.58%,and 8.23%,respectively. At the same time,the false alarm rate of SAE-DBN model can be controlled within 6%,which has a good prediction effect on the surface quality defects of hot rolled strip.