Mixup Enhanced Linear Robust Explainable Model for Identifying Key Factors of Surface Defects in Strip Steel Manufacturing

The surface defect is a common defect in strip steel products, which can seriously affect the various properties of the strip steel. This paper aims to propose a key influencing factor identification indicators based on the idea of explainable artificial intelligent to assist engineers in locating key variables related to surface defects. In the practical application, the obtained data show the characteristic of imbalance, and the defect data severely overlaps with the normal data, which makes it difficult to fully extract defect-related information. To address these challenges, in this paper, a novel fine-gained defect label is designed for constructing the supervised defect-related model, meanwhile, a Mixup based strategy is established to improve the robustness of the proposed method.