A Dent Defect Identification and Quantization Algorithm of Buried Pipeline Via the IMU Data

Oil and gas pipelines cross various complex terrains. Gular internal inspections are key for their safe and efficient function. Recently, the Inertial Measurement Unit (IMU) has been widely applied in pipeline inspections for its low cost, high precision, and suitability for long distances. However, current research by scholars internationally and domestically primarily focuses on pipeline strain identification using inertial navigation. There is a lack of intelligent identification and quantitative analysis for the size of dents in long-distance oil and gas pipelines based on inertial navigation data. Addressing this gap, this paper introduces a quantitative dent identification method for pipelines using a CNN-LSTM-Attention hybrid neural network. Initially, IMU inspection data undergo preprocessing and dimensionality reduction via the Principal Component Analysis (PCA) algorithm. A deep learning model comprising a CNN-LSTM-Attention neural network is constructed, capable of autonomously learning and identifying pipeline dent characteristics, optimizing the learning rate, and generating non-linear output, thereby realizing quantitative dent detection. The findings indicate that the proposed deep learning model possesses robust recognition and quantification capabilities, suitable for IMU data from pipeline internal inspections. The presented technical method effectively identifies and quantifies pipeline dent dimensions. Comparisons of convergence speeds at various learning rates revealed faster convergence at 0.001, with an accuracy of 92.4%. This allows for high-precision quantitative identification of pipeline dents, providing theoretical and technical support for the safety assessment of pipeline operations.