China Mechanical Engineering ›› 2023, Vol. 34 ›› Issue (23): 2854-2861.DOI: 10.3969/j.issn.1004-132X.2023.23.009

Previous Articles     Next Articles

Centrifugal Pump Fault Diagnosis Methods Based on Dislocation Superposition Methods and Improved Probabilistic Neural Networks

CHEN Jian1,2;XU Chang1,2;XU Tingliang1,2   

  1. 1.Institute of Noise and Vibration Engineering,Hefei University of Technology,Hefei,230009
    2.Automotive NVH Engineering & Technology Research Center Anhui Province,Hefei,230009
  • Online:2023-12-10 Published:2024-01-03

基于位错叠加法和改进概率神经网络的离心泵故障诊断方法

陈剑1,2;许畅1,2;徐庭亮1,2   

  1. 1.合肥工业大学噪声振动工程研究所,合肥,230009
    2.安徽省汽车NVH技术研究中心,合肥,230009
  • 作者简介:陈剑,男,1962年生,教授、博士研究生导师。研究方向为噪声振动控制、机械系统故障诊断。E-mail:hfgd8216@126.com。

Abstract:  Based on dislocation superposition method and improved probabilistic neural network, a fault diagnosis method of centrifugal pumps was proposed to solve the problems of online fault diagnosis using acoustic radiation signals of centrifugal pumps under strong background noise. Firstly, the acoustic radiation signals of centrifugal pumps were denoised by dislocation superposition method to enhance the fault information in acoustic radiation signals and improve the signal-to-noise ratio. The time domain features of acoustic signals were extracted to construct the time domain feature matrix. After dimensionality reduction of the obtained time domain feature matrix through principal component analysis, which was used as the inputs of machine learning probabilistic neural network. At the same time, Harris hawk optimization algorithm was used to optimize the parameters of the probabilistic neural network to get the diagnosis model, and then the improved probabilistic neural network was used to recognize the patterns of the centrifugal pump faults, and compared with a variety of diagnostic methods. The experimental results show that the dislocation superposition method may highlight the signal characteristics and realize signal enhancement, and the improved probabilistic neural network has a good ability of online fault diagnosis of centrifugal pump acoustic radiation signals. 

Key words: centrifugal pump, fault diagnosis, dislocation superposition method, probabilistic neural network, Harris hawk optimization algorithm

摘要: 提出了一种基于位错叠加法和改进概率神经网络的离心泵故障诊断方法以解决现场强背景噪声下基于离心泵声辐射信号的在线故障诊断问题。首先利用位错叠加法对采集的离心泵声辐射信号进行降噪处理,增强声辐射信号中的故障信息,提高信噪比;然后提取声信号时域特征以构造时域特征矩阵,通过主成分分析法对获得的时域特征矩阵进行降维处理,将降维后的信号作为机器学习概率神经网络的输入;同时用哈里斯鹰优化算法来优化概率神经网络参数得到诊断模型,继而用改进的概率神经网络对离心泵故障进行模式识别,并与多种诊断方法进行比较。实验结果表明:位错叠加法能够突出信号特征、实现信号增强,改进的概率神经网络具有良好的离心泵声辐射信号在线故障诊断能力。

关键词: 离心泵, 故障诊断, 位错叠加法, 概率神经网络, 哈里斯鹰优化算法

CLC Number: