Automated detection of gear tooth flank surface integrity: A cascade detection approach using machine vision

The surface integrity of gear tooth flanks significantly impacts the efficiency and reliability of gear transmission systems. This article proposes a cascaded detection approach using machine vision to comprehensively localize and identify thermal damages on tooth flanks following the grinding process. This method utilized an image enhancement to correct non-uniform illumination and a saliency detection based on the spectral residual algorithm to extract individual tooth flanks. Additionally, an image semantic segmentation model, GBSU-Net, was put forward to detect thermal damage regions on the tooth flank and quantify the severity of grinding burn with the area ratio. The experimental results demonstrated the efficacy of the proposed method on the gear surface image dataset, with the Dice coefficient and IoU metrics achieving 84.29 % and 73.95 %, respectively. The proposed method is applicable for real-time detection during gear machining processes because of its swift and accurate detection capability.