燃气切断阀故障状态监测及诊断技术研究
Research on Fault State Monitoring and Diagnosis Technology for Gas Shut-Off Valves

燃气切断阀作为燃气系统的核心组件之一，其潜在故障可能对系统安全性产生严重影响。为解决传统故障特征提取方法抗干扰能力弱、诊断准确率低的问题，本文提出一种基于互补集合模态分解(CEEMD)、核主成分分析(KPCA)、遗传算法优化支持向量机(GA-SVM)与D-S证据理论融合的燃气切断阀故障诊断方法。首先，通过CEEMD对燃气切断阀的故障信号进行分解，得到多个本征模态函数(IMF)分量，实现故障信号与噪声的初步分离；其次，利用KPCA对分解后的IMF分量进行特征提取与降维，构建高辨识度的多维度故障特征集，剔除冗余信息；然后，采用GA-SVM对优化后的特征集进行初步故障识别，通过遗传算法优化SVM的核函数参数与惩罚因子，提升模型的初步诊断性能；最后，引入D-S证据理论对GA-SVM的初步诊断结果进行多源信息融合，修正单一模型的诊断偏差。为验证该方法的有效性，以故障诊断的准确率、误判率作为评价指标，采用燃气切断阀模拟故障实验平台采集的正常信号及泄漏、卡滞、误动作等故障信号进行验证。实验结果表明，与传统方法相比，所提融合诊断方法的故障诊断准确率达到81.5%，误判率降低了23%~37%，显著提升了故障识别的稳定性与可靠性。该方法可为燃气切断阀的故障预警与精准诊断提供技术支撑，对保障燃气系统的安全稳定运行具有重要的工程应用价值。
As one of the core components of the gas system, the potential failure of the gas shut-off valve may have a serious impact on the safety of the system. To solve the problems of weak anti-interference ability and low diagnostic accuracy of traditional fault feature extraction methods, this paper proposes a gas shut-off valve fault diagnosis method based on complementary set mode decomposition (CEEMD), kernel principal component analysis (KPCA), genetic algorithm optimized support vector machine (GA-SVM), and D-S evidence theory fusion. Firstly, by decomposing the fault signal of the gas shut-off valve through CEEMD, multiple intrinsic mode function (IMF) components are obtained to achieve preliminary separation between the fault signal and noise; secondly, KPCA is used to extract and reduce the dimensionality of the decomposed IMF components, constructing a highly recognizable multi-dimensional fault feature set and removing redundant information; then, GA-SVM is used to perform preliminary fault identification on the optimized feature set, and genetic algorithm is used to optimize the kernel function parameters and penalty factors of SVM to improve the preliminary diagnostic performance of the model; finally, D-S evidence theory is introduced to fuse the preliminary diagnosis results of GA-SVM with multi-source information to correct the diagnosis bias of single model. To verify the effectiveness of this method, the accuracy and misjudgment rate of fault diagnosis were used as evaluation indicators. The normal signals and fault signals, such as leakage, jamming, and misoperation, collected from the gas cut-off valve simulation fault experimental platform were used for verification. The experimental results show that compared with traditional methods, the proposed fusion diagnosis method achieves a fault diagnosis accuracy of 81.5%,