基于深度学习的水电厂压油装置油泵振动异常状态检测  

Detection of abnormal vibration state of oil pump in hydraulic pressure device of hydropower plant based on deep learning

作　　者：高菘 付恩狄 柳本林 莫理 张龙浩 Gao Song;Fu Endi;Liu Benlin;Mo Li;Zhang Longhao(Western Maintenance and Testing Branch,Southern Power Grid Peak Shaving and Frequency Modulation Power Generation Co.,Ltd.,Guizhou Xingyi,562499,China)

机构地区：[1]南方电网调峰调频发电有限公司西部检修试验分公司,贵州兴义562499

出　　处：《机械设计与制造工程》2024年第10期83-87,共5页Machine Design and Manufacturing Engineering

摘　　要：提出一种基于深度学习算法的水电厂压油装置油泵振动异常状态检测方法。该方法使用低通滤波器对油泵振动信号进行双向平滑滤波处理,对处理后的信号进行VMD分解,将其分解为有限带宽模态分量。结合有限带宽模态分量的傅里叶变换结果,计算信号功率谱密度,并以振动信号标准差、偏度以及峰度作为时域特征参数,提取其特征。构建LSTM网络,遍历振动信号时频特征参数,得到前向层及后向层的输出序列,进行级联正向操作及反向操作,实现油泵振动异常状态检测。实验对比结果表明,所提方法检测精度较高、用时较短,是较为理想的水电厂压油装置油泵振动异常状态检测方法。A deep learning algorithm based method for detecting abnormal vibration states of oil pumps in hydraulic pressure equipment of hydropower plants is proposed.This method uses a low-pass filter to perform bidirectional smoothing filtering on the oil pump vibration signal,and decomposes the processed signal into finite bandwidth modal components through VMD decomposition.Based on the Fourier transform results of finite bandwidth modal components,it calculates the signal power spectral density,and extracts its features using the vibration signal standard deviation,skewness,and kurtosis as time-domain characteristic parameters.It constructs an LSTM network,traverses the time-frequency characteristic parameters of the vibration signal,obtains the output sequences of the forward and backward layers,performs cascaded forward and backward operations,and achieves abnormal state detection of oil pump vibration.The experimental comparison results show that the proposed method has high detection accuracy and short time,and is an ideal method for detecting abnormal vibration states of oil pumps in hydraulic power plants.

关 键 词：深度学习 水电厂 压油装置 油泵振动异常检测 变分模态分解 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]