An Effective Pipeline Vibration Trend Prediction Method Based on Adaptive Chirp Mode Decomposition and Long Short-Term Neural Networks

Pipelines are widely present in nuclear power plants, and their structural integrity is crucial for ensuring the safety of the plants. Traditional pipeline vibration control strategy is to measure vibrations during hot function tests to identify high-vibration pipelines for remediation. However, this method does not completely resolve the issue of pipeline vibrations, especially for high-vibration pipelines that are difficult with remediation. Therefore, a Structural Health Monitoring (SHM) system is employed for the identification and assessment of structure health. For the early identification of potential failures, the prediction of pipeline vibrations and the implementation of preventive measures is important. In view of the multi-frequency, modal aliasing, and high noise characteristics of pipeline vibration signals, a method that balances efficiency and accuracy, with good adaptability, is needed for online continuous prediction. This paper introduces a novel vibration trend prediction method that combines the adaptive chirp mode decomposition (ACMD) method and the long short-term memory neural network (LSTM). This approach effectively predicts future vibration signals based on historical data. Finally, the effectiveness of the method is demonstrated through both numerical simulations and field experiments.