Thickness Measurement and Anomaly Detection in Transmission Belts using Adaptive Dual-Scale Analysis

Abstract

Abnormal variations in transmission belt thickness are a critical cause of mechanical transmission failure, efficiency reduction, and abnormal noise, necessitating highly accurate and automated detection methods. With technological advancements, 3D point cloud detection has emerged as a superior solution, overcoming the limitations of 2D imaging by enabling precise defect quantification and demonstrating strong robustness in complex working conditions. This paper proposes a novel automated method for detecting abnormal thickness mutations in transmission belts. High-density 3D point cloud data are acquired using a line laser sensor and preprocessed for analysis. To improve computational efficiency, a uniform slicing method is employed to reduce data dimensionality, while reference baselines are established along both edges of the pulley to provide geometric benchmarks for thickness calculation. To address interference from industrial markings and noise, an adaptive window adjustment strategy combined with curvature constraints is applied to extract feature points on the belt’s backside, and a neighborhood consistency-weighted model is designed to refine thickness computation. Furthermore, to account for inherent thickness non-uniformity in transmission belts, a dual-scale median difference analysis and dynamic thresholding mechanism are proposed, enabling cooperative detection of local mutations and global trends for anomaly identification. Experimental results demonstrate that the proposed method achieves a maximum absolute thickness measurement error of 0.029 mm, a repeatability error of ±0.01 mm, and an average detection time of 0.65 s. The anomaly detection accuracy reaches 94.4%, with an average processing time of 2.09 s. The method exhibits high robustness and measurement precision, meeting practical inspection requirements and providing robust technical support for transmission belt maintenance and quality control.