Related Experiment Video
Updated: Jan 9, 2026
05:11
High-precision Electromagnetic Flowmeter with Empty Pipe Detection via Complex Programmable Logic Device-based Waveform Recognition
Published on: June 27, 2025
596
Pipeline Leakage Identification Based on Acoustic Sensors and EPSO-1D-CNN-Bi-LSTM Model
Niannian Wang1,2,3, Kuankuan Zhang1,2,3, Xingyi Wang1,4
1Yellow River Laboratory, Zhengzhou University, Zhengzhou 450001, China.
Sensors (Basel, Switzerland)
|December 11, 2025
Summary
This study introduces an advanced AI model for detecting and classifying water pipe leaks. The one-dimensional convolutional neural network and bidirectional long short-term memory network fusion model (1D-CNN-Bi-LSTM) achieves high accuracy, improving urban water system management.
Area of Science:
- Civil Engineering
- Artificial Intelligence
- Water Resource Management
Background:
- Urban water supply systems face significant challenges with underground pipe leakage.
- Current in-pipe inspection methods for leak detection have limitations in accuracy, rely on experienced staff, and pose health risks.
- Accurate leakage detection and severity classification are critical for maintaining pipeline integrity and water supply efficiency.
Purpose of the Study:
- To develop and validate an advanced AI-driven model for accurate water pipe leakage detection and severity classification.
- To enhance the model's performance through hyperparameter optimization and multi-feature data fusion.
- To provide a more reliable and effective alternative to traditional in-pipe inspection methods.
Main Methods:
- Development of a hybrid deep learning model: one-dimensional convolutional neural network and bidirectional long short-term memory network (1D-CNN-Bi-LSTM).
- Optimization of model hyperparameters using an enhanced particle swarm optimization (EPSO) algorithm.
- Implementation of multi-feature fusion for improved data representation and model robustness.
- Validation through ablation experiments and full-scale field tests.
Main Results:
- The 1D-CNN-Bi-LSTM model achieved 98.33% accuracy in both leakage detection and severity classification.
- Ablation studies confirmed the significant contributions of the EPSO algorithm and Bi-LSTM modules.
- The model demonstrated strong anti-noise capabilities and stable recognition performance in real-world conditions.
- The proposed method significantly outperformed existing models in accuracy and reliability.
Conclusions:
- The developed 1D-CNN-Bi-LSTM model offers a highly accurate and effective solution for pipeline leakage detection and severity assessment.
- This AI-driven approach reduces the reliance on potentially hazardous and less accurate in-pipe inspection devices.
- The findings pave the way for improved management and maintenance of urban water supply networks.
