Non-destructive monitoring of forming quality of self-piercing riveting via a lightweight deep learning

Abstract

Self-piercing riveting (SPR) has been widely used in automobile body jointing. However, the riveting process is prone to various forming quality failures, such as empty riveting, repeated riveting, substrate cracking, and other riveting defects. This paper combines deep learning algorithms to achieve non-contact monitoring of SPR forming quality. And a lightweight convolutional neural network with higher accuracy and less computational effort is designed. The ablation and comparative experiments results show that the lightweight convolutional neural network proposed in this paper achieves improved accuracy and reduced computational complexity. Compared with the original algorithm, the algorithm’s accuracy in this paper is increased by 4.5$$\%$$ % , and the recall is increased by 1.4$$\%$$ % . In addition, the amount of redundant parameters is reduced by 86.5$$\%$$ % , and the amount of computation is reduced by 47.33$$\%$$ % . This method can effectively overcome the limitations of low efficiency, high work intensity, and easy leakage of manual visual inspection methods and provide a more efficient solution for monitoring the quality of SPR forming quality.