Research On Prediction Of Multi-class Surface Defects Of Hot Rolled Strip Based On Data-driven

The digital industrial upgrading of the steel industry is an important component of China’s industrial economy moving towards a digital economy during the 14 th Five Year Plan period.How to fully utilize the digital information of the hot rolling production process in the new era is of great significance for improving the surface quality of hot rolled strip products and catering to the national "dual carbon" strategy.Based on the national steel industry development strategy and the development status of steel industry informatization,this paper uses the obtained hot rolling production process quality data set,uses data mining,feature engineering and other methods to design an interpretable collaborative prediction method for multiple types of surface defects of hot rolled strip steel,to help the quality control of hot rolled surface defects.The main content of this article is as follows:(1)In order to ensure the interpretability of input variables and improve the classification and prediction accuracy of the model,the formation mechanism and related process variables of various hot rolled surface defects are sorted out,and the combination of filtering,embedding,and packaging feature selection is explored to reduce the dimensionality of input variables.Drawing on the characteristics of LightGBM algorithm and REFCV,a training method for hot rolled surface defect prediction and classification model that can obtain defect related feature weight sorting is designed;(2)In order to alleviate the influence of the imbalance of positive and negative samples in the original hot rolling surface quality data set on the classification accuracy of the prediction model,the proportion of positive and negative samples in the training data set was adjusted by combining the methods of removing abnormal and out of limit samples,synthesizing minority samples,and selecting classification threshold;The characteristics of the GBDT-LR model and the advantages of its application in the online prediction scenario of hot rolled strip steel quality were elaborated.The effectiveness of this method was verified through the classification and prediction of edge defects and iron oxide skin pressing defects;(3)In order to make full use of the prediction information of online samples by multi class defect classifiers,and considering the association relationship of multi class defects,the spectral clustering and collaborative filtering algorithm are applied to the collaborative prediction of the probability of multi class defects.Combined with the historical prediction information,the abnormal prediction results are balanced to improve the reliability of the classification prediction of real-time online samples.