Research on a Novel Unsupervised-Learning-Based Pipeline Leak Detection Method Based on Temporal Kolmogorov–Arnold Network with Autoencoder Integration

Sensors, Journal Year: 2025, Volume and Issue: 25(2), P. 384 - 384

Published: Jan. 10, 2025

Artificial intelligence (AI) technologies have been widely applied to the automated detection of pipeline leaks. However, traditional AI methods still face significant challenges in effectively detecting complete leak process. Furthermore, deployment cost such models has increased substantially due use GPU-trained neural networks recent years. In this study, we propose a novel detector, which includes new model and sequence labeling method that integrates prior knowledge with reconstruction error theory. The proposed combines Kolmogorov-Arnold Network (KAN) an autoencoder (AE). This (AE), forming hybrid framework captures complex temporal dependencies data while exhibiting strong pattern modeling capabilities. To improve detection, developed unsupervised anomaly based on theory, incorporates in-depth analysis curve along knowledge. significantly enhances interpretability accuracy Field experiments were conducted real urban water supply pipelines, benchmark dataset was established evaluate against commonly used methods. experimental results demonstrate achieved high segment-wise precision 93.1%. Overall, study presents transparent robust solution for facilitating large-scale, cost-effective development digital twin systems emergency management.

Language: Английский

A Comparative Analysis of Pipeline Leakage Detection Using Different Ml Techniques

Published: Jan. 1, 2025

Language: Английский

Pipeline and Rotating Pump Condition Monitoring Based on Sound Vibration Feature-Level Fusion

Machines, Journal Year: 2024, Volume and Issue: 12(12), P. 921 - 921

Published: Dec. 16, 2024

The rotating pump of pipelines are susceptible to damage based on extended operations in a complex environment high temperature and pressure, which leads abnormal vibrations noises. Currently, the method for detecting conditions pumps primarily involves identifying their sounds vibrations. Due background noise, performance condition monitoring is unsatisfactory. To overcome this issue, pipeline proposed by extracting fusing sound vibration features different ways. Firstly, hand-crafted feature set established from two aspects vibration. Moreover, convolutional neural network (CNN)-derived one-dimensional CNN (1D CNN). For CNN-derived sets, selection presented significant ranking according importance, calculated ReliefF random forest score. Finally, applied at level. According signals obtained experimental platform, was evaluated, showing an average accuracy 93.27% conditions. effectiveness superiority manifested through comparison ablation experiments.

Language: Английский