基于PCA-RF-BP的液压系统异常状态诊断策略  

Abnormal State Diagnosis Strategy of Hydraulic System Based on PCA-RF-BP

作　　者：董治国 贺虎 王晓晖[2,3] DONG Zhi-guo;HE Hu;WANG Xiao-hui(Changsha Pump Works Co.,Ltd.,Changsha 410000,China;College of Energy and Power Engineering,Lanzhou University of Technology,Lanzhou 730050,China;Doctor Innovation Station of Chaoda Valve Group Lishui Co.,Ltd.,Lishui 323000,China)

机构地区：[1]长沙水泵厂有限公司,湖南长沙410000 [2]兰州理工大学能动学院,甘肃兰州730050 [3]超达阀门集团丽水有限公司博士创新站,浙江丽水323000

出　　处：《液压气动与密封》2024年第9期104-111,共8页Hydraulics Pneumatics & Seals

摘　　要：液压系统状态监测信号受回路特性与泵机组运行工况影响,呈现复杂非线性且异常状态难以准确识别与诊断。为此,提出了一种基于主成分分析与随机森林BP神经网络(PCA-RF-BP)的液压系统异常状态诊断策略,用于提高设备监测系统的诊断效率。首先,基于状态监测数据进行主成分分析以降低数据维度,同时计算T 2和SPE统计量进行过程状态的实时异常检测;其次,采用随机森林BP神经网络对异常样本进行预测分类。实验结果表明,所提方法能够有效地诊断液压系统泄漏状态,检测延迟至多5个样本点,预测分类精度达到99.88%,相较于现有方法平均提高了4.63%。The state monitoring signal of the hydraulic system is affected by the characteristics of the circuit and the operating conditions of the pump unit,presenting complex and nonlinear characteristics,and the abnormal state is difficult to identify and diagnose accurately.This work proposes a diagnosis strategy for the abnormal state of the pump hydraulic system based on principal component analysis and random forest BP neural network to improve the diagnostic efficiency of equipment monitoring systems.Firstly,principal component analysis is performed based on state monitoring data to reduce data dimensions,while T 2 and SPE statistics are calculated for real-time anomaly detection of the process states.Secondly,a random forest BP neural network is used to predict and classify the abnormal samples.The experimental results show that the proposed method can effectively diagnose the leakage states of the system,detect a delay of 5 sample points,and achieve a prediction classification accuracy of 99.88%,which is an average improvement of 4.63%compared to existing methods.

关 键 词：泵 液压系统 异常检测 故障诊断 主成分分析 深度学习 

分 类 号：TH137[机械工程—机械制造及自动化]