Recognition of open-set appearance quality of resistance spot welding based on triplet loss

Le résumé fourni par la source

This study addresses the challenge of open-set recognition of resistance spot welding appearance quality by proposing a recognition method based on triplet loss. An experimental platform for resistance spot welding was established to construct a point cloud dataset comprising five categories of welded joints. The OFTFHC model was employed for feature extraction. The proposed method incorporates angular loss, dynamic anchor loss, and contrast loss to optimize the spatial distribution of features and enhance the model's capability to recognize unknown categories. Experimental results demonstrate that the model achieves satisfactory performance in open-set recognition, with an area under the receiver operating characteristic curve (AUROC) of up to 0.944 and an average precision (AP) of up to 0.769, effectively distinguishing between known and unknown category samples. This study offers a novel approach for the intelligent identification of resistance spot welding quality.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.