A Machine Learning-Based Framework for Image Quality Inspection of Automotive Metal Stamping Parts

—Traditional quality control for automotive metal stamping parts relies heavily on Checking Fixtures (C/Fs). These fixtures are custom-engineered for individual components, resulting in high initial design and manufacturing costs, limited versatility, and time-consuming processes for each new part. Moreover, C/F inspection is a manual and subjective process, which introduces human error, measurement variability, and slower throughput in high-volume production environments. This study proposes an automated imaging-based quality inspection framework utilizing a 3D laser scanner and the k-Nearest Neighbors (k-NN) machine learning algorithm. The framework systematically analyzes complex point cloud data of scanned parts through a structured sequence of steps: Data Acquisition, Segmentation, Pre-processing, Feature Recognition, Data Analysis, Post-processing, and Final Decision-making. To ensure both high accuracy and maximum speed, each step involves direct and immediate comparison with nominal Computer-Aided Design (CAD) data or a pre-established training set. The k-NN algorithm plays a central role in the analysis phase, effectively using Euclidean distances to distinguish noise from true features, recognize geometric elements such as holes, and reliably detect defects including material burrs and dimensional springback. The proposed system offers significant advantages over traditional C/Fs, including greater versatility across diverse component geometries and substantially reduced labor costs through full automation. Additionally, it ensures faster inspection times and consistent, objective accuracy, thereby eliminating the subjectivity, human error, and physical degradation associated with conventional fixtures. This automated framework represents a more sustainable, efficient, and robust quality control solution, aligning with the future needs of the automotive stamping industry.