Based On Data-driven Surface Defect Diagnosis Of Hot-rolled Strip

The surface quality defect of hot rolled strip is one of the important factors affecting the quality of strip.How to effectively reduce the probability of surface defects in the production process has become a hot topic in the field of rolling.In this paper,the prediction and tracing diagnosis models of surface defects of hot rolled strip were established by datadriven method.In this paper,the status of surface defect diagnosis of rolled strip,the development status of data-driven and the application.In view of the common defects of hot rolled strip edge warping,iron sheet printing and burr,the associated data of relevant defects were sorted out and used as the data source of subsequent data model.Based on the deep belief network,a network model was established which integrates each dimension reduction layer,and the data relationship between pre-test factors related to defects and specific defects was established,the defect prediction of the final product surface quality could be realized by the fluctuation of the pretest parameters,and the overall prediction accuracy could reach more than 80% by using the actual field data.Based on the surface defect prediction model of hot rolled strip.A fault tree method with contribution factor was designed to trace defects,and compared with the common contribution graph method and the traditional fault tree method,the fault tree method with contribution factor could locate the fault source more quickly and accurately,and was consistent with the field results.On the basis of predicting and tracing surface defects of hot rolled strip.The hot-rolled strip surface defect diagnosis system was designed based on industrial field measurement data source,computer hardware system,LabVIEW virtual simulation platform and MATLAB mathematical calculation software,the data input,defect prediction,traceability and information storage were realized.