The Application of Random Forest Model Based on Acoustic Emission Parameters in Valve Internal Leakage Detection

针对天然气站场阀门内漏检测效率低、环境干扰强的问题等问题，本研究提出基于声发射参数与随机森林算法的阀门内漏检测方法。通过构建气体阀门内漏模拟实验平台，在0.3~0.8 MPa压力范围内采集DN100闸阀的声发射信号，采用小波包分解对18.75~75.0 kHz特征频段进行降噪处理，提取压力、幅度、能量、RMS、ASL等5个特征参数构建特征矩阵。通过随机森林分类模型对阀门正常关闭(标签-1)与内漏状态(标签1)进行分类建模，结果表明：在测试集上，模型准确率可达93.33%，其中阀门压差与能量参数对阀门内漏判别具有较高贡献度。该方法有效解决了传统检测方法对微小泄漏响应迟滞的问题，为工业阀门状态监测提供了新的技术路径。Aiming at the problems of low detection efficiency and strong environmental interference of valve internal leakage in natural gas station, this study proposes a valve internal leakage detection method based on acoustic emission parameters and random forest algorithm. By constructing a gas valve internal leakage simulation experiment platform, the acoustic emission signals of DN100 gate valve were collected in the pressure range of 0.3~0.8 MPa. The wavelet packet decomposition was used to denoise the characteristic frequency band of 18.75~75.0 kHz, and five characteristic parameters such as pressure, amplitude, energy, RMS and ASL were extracted to construct the characteristic matrix. The random forest classification model is used to classify and model the valve normal closing (label-1) and internal leakage state (label 1). The results show that the accuracy of the model can reach 93.33% on the test set, and the valve pressure difference and energy parameters have a high contribution to the valve internal leakage discrimination. This method effectively solves the problem of hysteresis response of traditional detection methods to small leakage, and provides a new technical path for industrial valve condition monitoring.