可靠性强化试验技术在电动闸阀故障识别中的应用  

Application of reliability enhancement test technology in fault identification of electric gate valve

作　　者：张林[1,2] 汤文斌 刘杰 闫晓[1] 湛力[1] 李明刚[1] 周吴[2] ZHANG Lin;TANG Wenbin;LIU Jie;YAN Xiao;ZHAN Li;LI Minggang;ZHOU Wu(Nuclear Power Institute of China,Chengdu 610213,China;School of Mechanical and Electrical Engineering,University of Electronic Science and Technology of China,Chengdu 611731,China;College of Computer Science,Sichuan University,Chengdu 610065,China)

机构地区：[1]中国核动力研究设计院,成都610213 [2]电子科技大学机械与电气工程学院,成都611731 [3]四川大学计算机学院,成都610065

出　　处：《流体机械》2025年第3期110-118,共9页Fluid Machinery

基　　金：国家自然科学基金项目(52475120)。

摘　　要：为探究电动闸阀的可靠性及轻微故障识别技术的难点,采用可靠性强化试验技术在非破坏条件下开展电动阀门故障模拟试验,成功获取了2种典型故障模式的响应数据,最后采用决策树、随机森林和梯度提升3种机器学习方法,并结合网格搜索和Actor-Critic强化学习2种优化方案,对试验数据进行了分析。结果表明,可靠性强化试验技术能够在不破坏阀门的前提下,低成本、高效地模拟阀门卡滞和卡死故障;电流与振动特征的组合显著提高了故障识别精度,尤其是在随机森林和梯度提升算法中正常数据识别精度达98%,在随机森林和决策树算法中故障数据识别精度最高达到82%;多特征融合有助于充分利用数据的特性提升模型性能;强化学习在提升随机森林性能方面表现出色,振动特征下准确率提升21%,振动电流组合特征下提升15%,电流特征下提升6%;但在梯度提升树中未能显著改善性能,反而导致其产生一定程度的下降;决策树适合处理多特征组合数据,表现出更高的故障识别精度;而随机森林在单一特征和多特征组合下均表现稳定,训练时间和识别精度均具有较高的鲁棒性;梯度提升树在某些特征类型下的表现波动较大,应针对其特点进行进一步优化;在实际工业故障诊断中,应优先考虑随机森林和决策树算法,结合多特征融合和优化方法以提升分类性能。研究可为电动闸阀的故障识别提供参考。To investigate the reliability of electric gate valves and challenges in minor fault identification,reliability enhancement testing(RET)technology was applied to simulate valve faults under non‑destructive conditions.Response data from two typical failure modes were collected and analyzed using three machine learning methods(decision tree,random forest,gradient boosting)combined with two optimization approaches(grid search,actor‑critic reinforcement learning).Results indicate that RET effectively simulates valve sticking and jamming faults without physical damage,achieving cost‑efficient fault replication.The integration of current and vibration features significantly enhances fault recognition accuracy:normal data identification accuracy reaches 98%with random forest and gradient boosting,while fault data identification peaks at 82%using random forest and decision tree.Multi‑feature fusion improves model performance by leveraging data characteristics.Reinforcement learning boosts random forest accuracy by 21%(vibration features),15%(vibration‑current features),and 6%(current features),but degrades gradient boosting performance.Decision Tree excels in processing multi‑feature data with superior fault recognition,while random forest maintains robust performance across single and combined features.Gradient boosting exhibits instability requiring feature‑specific optimization.For industrial applications,random forest and decision tree with multi‑feature fusion are recommended to enhance classification performance.This study provides technical references for electric gate valve fault diagnosis.

关 键 词：电动阀门 可靠性强化 故障模拟 故障识别 

分 类 号：TL353[核科学技术—核技术及应用] TP183[自动化与计算机技术—控制理论与控制工程]