The main circuit of the reactor has been running for a long time in a high-temperature, high-pressure and complex load environment. Under the action of corrosion, erosion, fatigue and other factors, cracks may occur and further develop into leakage accidents, which poses a threat to the safe and stable operation of nuclear power plants. In response to the problem that the traditional monitoring method has limited ability to characterize the early leakage state, this paper carries out a one-loop main channel state monitoring research based on leakage sound signals and machine learning. By analyzing the time-frequency response characteristics of leakage sound signals under different pressure conditions, the mapping relationship between acoustic characteristics and pipeline operation pressure is established. Build a dual-sensor pipeline leakage sound signal collection system, carry out experiments under five pressure conditions of 0.2 to 1.0MPa, and extract the mean square root value, peak, steepness, power spectrum density, peak frequency, spectral entropy and frequency band energy and other characteristics from dimensions such as time domain, frequency domain and frequency band energy. The results show that changes in pipeline pressure can cause significant changes in the amplitude, energy and spectrum distribution of the leakage sound signal, among which the energy-related characteristics show high sensitivity to pressure changes. Furthermore, the sliding window was used to construct 1,510 groups of samples, and a random forest pressure prediction model was established by integrating the multi-dimensional acoustic characteristics of the dual test point. The test set decision coefficient R2 is 0.916, the root mean square error is 0.079MPa, and the average absolute error is 0.053MPa; the cross-verification average R2 is 0.921, and the standard deviation is 0.013, indicating that the model has good prediction accuracy and stability. Finally, the SHAP method is used to analyze the characteristic contribution, and the results show that energy-related characteristics such as the root mean square value of the double test point have a high contribution to pressure prediction. This paper realizes the expansion from the characteristic analysis of the leakage sound signal to the quantitative characterization of pipeline operation pressure, which can provide method support for the monitoring of the leakage status of the main channel in one loop and the subsequent identification of the degree of leakage under different leakage sizes and intelligent early warning.

One-loop main channel; pipeline leakage; sound signal; time-frequency characteristics; random forest; SHAP

暂无