Fault Location of Strip Steel Surface Quality Defects on Hot-Rolling Production Line Based on Information Fusion of Historical Cases and Process Data

Abstract

Surface quality is the most important index to improve the overall quality of strip steel. In order to implement the fault location on the hot-rolling line with surface defects of strip steel, a fault tracing model based on information fusion of historical production cases and process data is proposed. For historical cases, the model determines the defect cause labels through text similarity calculation, and fuzzy semantic inference is used to obtain the probability distribution of defect causes on this basis; for the process data, the model uses L1 regularization method for feature selection, and XGBoost integration method is used to train the correlation model between process data and defects to determine the contribution of each feature in the data source. Finally, based on the D-S evidence theory, different rules are set to merge the two judgments to determine the probability of each source of failure on the hot-rolling production line. The model is applied to the real production environment of iron and steel enterprises, and it is verified that the proposed method can effectively assist experts in decision-making, which greatly improves the efficiency of tracing the source of faults on the hot-rolling production line.

Research on a Classification Method for Strip Steel Surface Defects Based on Knowledge Distillation and a Self-Adaptive Residual Shrinkage Network

“…Li et al [15] used the multiscale information feature fusion method to re-extract the texture information of the bottom layer of the image to focus more attention upon the slender and easily overlooked defects of the strip steel surface, and the detection accuracy of this method reached 98.26%. Wang et al [16] fused the historical data of strip defects to track model faults, which can assist expert decision-making.…”

“…The method of [13] can combat uneven samples, but it ignores the feature extraction ability of backbone networks and the efficiency and deployment of the models. Therefore, inspired by references [14][15][16], this pa-per integrates fine-grained information extraction, data sample enhancement, unbalanced sample confrontation, and model generalization performance transfer.…”