Zero-error-production through inline-quality control of presshardened automotive parts by multi-camera systems

Abstract

Abstract Visual inspection often represents a bottleneck in the production chain due to limited receptivity and human fatigue. Depending on the inspection task, human classification error decisions greater than 10% are not uncommon. In order to make the manufacturing process more robust and sustainable, this paper presents innovative automated inline monitoring to realize a zero-error strategy in the field of industrial manufacturing. The central idea is to detect surface defects in the running production process by data fusion of a multi-camera system. Due to the required data analysis in conjunction with the targeted high throughput rates in production, this generates large amounts of data to process. Massively parallel hardware structures are used to enable defect detection within the production cycle. The paper especially addresses how to tackle boundary conditions in productive press-lines like the software-based mitigation of placement tolerances of parts and real-time capabilities. As a result, benefits could be achieved in terms of minimum detectable failure sizes and inspection speed, enabling 100% inline inspection of produced parts. The feasibility of the presented approach is demonstrated on realistic press-hardened deep-drawn automotive parts.