关键词: 激光点云数据, 隧道裂缝探测, 联邦加权学习算法, 识别准确度, 算法性能对比

Abstract: The three-dimensional laser scanners for detecting tunnel cracks have disadvantages such as low recognition accuracy and poor anti-interference ability. Therefore, a new research approach for crack detection based on federated weighted learning algorithm is proposed. Based on tunnel laser point cloud data, an optimized federated weighted learning algorithm is employed, and asynchronous and residual testing adaptive adjustment algorithms are adopted to achieve overall accurate detection of tunnel cracks. Experiments are conducted in the Linyi-Tengzhou expressway tunnel, focusing on several indicators such as reliability, accuracy, and measurement accuracy of crack detection. The proposed algorithm is compared with traditional ones. The results show that the proposed method can effectively improve the reliability and accuracy of tunnel crack detection, exhibiting good performance in detecting crack width accuracy. When interference factors such as dust and exposed steel bars appear in the detection results, the proposed algorithm still exhibits significant advantages in reliability compared to traditional algorithms, achieving an accuracy of over 95% and a misidentification rate of less than 10%, thus ensuring the robustness of the algorithm′s application effect. Through an on-site engineering practice, the minimum deviation between the crack width identified by the proposed algorithm and the manually measured value is only 0.06 mm, verifying its good crack recognition accuracy.

Key words: laser point cloud data, tunnel crack detection, federated weighted learning algorithm, identification accuracy, comparison of algorithm performance