LiDAR-Based Crack and Fault Detection System for Pipelines

Pipeline integrity is essential for ensuring safe fluid transport, with cracks posing a major risk to operational safety and environmental protection. Traditional inspection methods are often expensive and time-intensive. To address this, we present a Lidar-based crack detection system mounted on a servo motor, which continuously captures distance and angle data from the inner surface of pipelines. We employ both unsupervised and supervised machine learning approaches to detect anomalies. Isolation Forest is used to identify potential cracks in an unsupervised manner, allowing the detection of unknown anomalies. In parallel, the XGBoost algorithm is applied to classify known cracks using labelled data. Our experimental results demonstrate that the Isolation Forest achieves an anomaly detection rate of 96%, while XGBoost attains a classification accuracy of 89%. This dual approach enhances the robustness and efficiency of pipeline crack detection, offering a reliable alternative to conventional inspection methods. A mobile application is also developed to view the position, angular position and the intensity (depth) of the crack.