Real-time quality classification of robot self-piercing rivet processes in CFRP-steel joints using machine learning

Abstract

This study presents a machine learning-based method for real-time quality classification of robotic self-piercing rivet (SPR) processes using pressing force as an input. Unlike previous studies focusing on general SPR processes, this study specifically targets automated robot SPR processes. Data for the study was acquired using a load cell, strain gauge, and accelerometer to develop a nondestructive quality evaluation system for real-time robotic SPR processes. This study introduced an effective and robust data preprocessing method to enhance quality classification by applying signal processing and data dimensionality reduction techniques to pressing force data. While recent advances in deep learning and machine learning have explored various complex networks to improve performance, research specifically addressing the quality of automated robot SPR processes remains limited. Although robotic SPR quality classification may appear similar to conventional SPR quality assessment, the data input differs significantly owing to the vibrations generated by industrial robots. These vibrations reduce the effectiveness of conventional classification methods, making accurate quality assessment more challenging. To address these challenges, six relatively simple yet effective machine learning algorithms were employed to demonstrate the advantages of the preprocessed features of the robotic SPR process, and the optimal algorithm was selected by comparing these algorithms. The proposed method was validated using real SPR data from two automotive production lines. By combining time-domain, processed, and dimensional reduced features from pressing force data, the machine learning model achieved 99.6 % accuracy in real-time SPR quality classification. This confirms its effectiveness and reliability for use in robotic manufacturing environments.