Principal component analysis-based image fusion routine with application to automotive stamping split detection

This study discusses the development and implementation of noncontact split detection method, for automotive stamping press lines. The system features a novel fusion routine that combines thermal and visible images in real-time, assisted with principle component analysis (PCA) subroutine. The thermal detector scans the temperature maps of the highly reflective steel sheets in the die cavity to locate abnormal temperature readings that might be indicative of high local wrinkling pressure, while the visible vision system offsets the blurring effect caused by heat diffusion across the surface and provide a spatial reference. The employed PCA uses a new singular value decomposition (SVD) that is more efficient than standard SVD computations, enabling the PCA to be applied in real-time acquisitions (∼30Hz). The PCA affects the images by reducing the nonvalue data content (reduce redundancy, noise) while highlighting important features. The fusion is done using a pixel-level algorithm using different variations, where each is assessed for performance. The proposed detection system has been tested on an automotive pressline to assess the formability of complex-shaped panels and have shown high detection success rate. Different splits with variant shape, size, and severity have been detected under actual operating conditions.