Research on online detection method of bearing outer surface defects based on single station

Purpose To address the industrial need for efficient and high-precision quality inspection, this study proposes an intelligent online detection method for bearing outer surface defects based on a single-station setup. Design/methodology/approach The method aims to significantly enhance detection accuracy and speed while reducing the system's footprint and cost. Its core innovation is a novel image acquisition system that employs prism reflection and a single-line scanning visual sensor, enabling the real-time capture of the bearing's entire cylindrical surface. A two-stage detection strategy integrating preliminary screening and follow-up confirmation is developed. In this strategy, the bearing surface is first segmented into multiple longitudinal dimensions. Suspected defect areas within each dimension are then precisely extracted and subjected to a dedicated recurring detection process, powered by a newly constructed deep learning network model. Three key algorithms – for dimension segmentation, region extraction and recurring detection – are seamlessly integrated and executed in a parallel computing architecture to maximize processing throughput. Findings Experimental results demonstrate that the proposed method achieves marked improvements in both detection accuracy and processing speed while effectively minimizing missed detections. Originality/value The primary contribution is an integrated optical-computational framework that delivers reliable, high-performance inline inspection with a simplified and cost-effective hardware solution, fully meeting the practical demands of automated industrial production lines.