Intelligent Identification of Internal Leakage of Spring Full-Lift Safety Valve Based on Improved Convolutional
Shuxun Li1,2, Kang Yuan1,2, Jianjun Hou1,2
1School of Petrochemical Technology, Lanzhou University of Technology, Lanzhou 730050, China.
Sensors (Basel, Switzerland)
|September 13, 2025
Summary
This study introduces an improved convolutional neural network (CNN) for accurate safety valve internal leakage identification using acoustic emission signals. The novel method achieves 99.7% accuracy, enhancing industrial safety and reducing economic losses.
Area of Science:
- Mechanical Engineering
- Signal Processing
- Artificial Intelligence
Background:
- Spring full-lift safety valves are crucial for pressure relief systems but prone to internal leakage due to valve seat damage, posing safety risks and economic losses.
- Existing fault diagnosis methods, like model-based approaches, face parameter determination challenges, while data-driven methods like convolutional neural networks (CNNs) require experienced hyperparameter tuning and struggle with temporal and multi-scale feature extraction for valve leakage identification.
Purpose of the Study:
- To develop an effective and accurate method for identifying the internal leakage state of safety valves.
- To address the limitations of existing CNN models in capturing temporal and multi-scale features and in hyperparameter optimization for valve fault diagnosis.
Main Methods:
- Acquisition of acoustic emission signals from safety valves in different internal leakage states using a high-frequency FPGA system.
- Processing signals into 2D time-frequency diagrams and inputting them into an improved CNN model incorporating Leaky Rectified Linear Unit (LReLU), random pooling, horned lizard optimization algorithm (HLOA) for hyperparameter tuning, Bidirectional Gated Recurrent Unit (BiGRU), and Selective Kernel Attention Module (SKAM).
Main Results:
- The proposed improved CNN model achieved an average recognition accuracy of 99.7% for safety valve internal leakage states.
- The model demonstrated superior performance compared to existing models like ResNet-18 in identifying valve internal leakage.
Conclusions:
- The developed method offers an effective solution for diagnosing internal leakage in safety valves, significantly improving accuracy and reliability.
- The signal processing and diagnostic approach can be extended to fault diagnosis in other mechanical equipment, with future work focusing on lightweight networks and multi-source data fusion for enhanced real-time performance and robustness.
