基于多传感器信息融合和CNN-BIGRU-Attention模型的液压防水阀故障诊断方法  

Fault diagnosis method for hydraulic waterproof valves based on a multi-sensor information fusion and CNN-BIGRU-Attention model

作　　者：肖遥 向家伟[1] 汤何胜[1] 任燕[1] XIAO Yao;XIANG Jiawei;TANG Hesheng;REN Yan(School of Mechanical and Electrical Engineering,Wenzhou University,Wenzhou 325000,China)

机构地区：[1]温州大学机电工程学院,浙江温州325000

出　　处：《机电工程》2024年第9期1517-1528,共12页Journal of Mechanical & Electrical Engineering

基　　金：国家自然科学基金资助项目(52175060);浙江省自然科学基金资助项目(LY20E050028)。

摘　　要：在建筑工程领域,尤其是在工程搅拌设备中,涉及到复杂液压工作介质,液压防水阀往往会出现不同程度的故障。此外,恶劣的工作环境和复杂的噪声背景使得对液压防水阀的故障进行诊断变得困难。为了解决这一难题,提出了一种基于多传感器信息融合和卷积神经网络-双向门控循环单元-自注意力机制(CNN-BIGRU-Attention)模型的防水阀故障诊断方法。首先,考虑到单一传感器振动信号难以充分表达故障特征,该方法使用了3个传感器采集含噪声的振动信号,并进行了必要的预处理;其次,提取了信号的16个时域特征、5个频域特征以及3个时频域特征,并利用熵权法进行了特征融合,达到了增强特征的目的;然后,将融合的多维特征集输入到CNN-BIGRU-Attention模型中进行了特征识别;最后,利用实际的液压防水阀故障诊断实验,验证了该方法的有效性。研究结果表明:采用多传感器提取的特征更为全面,信息融合有助于捕捉更完整的隐藏数据特征,从而显著提高诊断的准确率;相较于其他特征识别方法,采用CNN-BIGRU-Attention模型的液压防水阀故障诊断准确率可分别提高6.7%、4.6%和14.2%,达到了96.86%,证明了该方法的有效性。该方法将先进的机器学习技术与实际工程应用相结合,为建筑工程问题提供了一种新颖、有效的解决方案。In the field of construction engineering,particularly with respect to mixing equipment in projects,the complexity of hydraulic working media often resulted in varying degrees of malfunction in hydraulic waterproof valves.Moreover,harsh working environments and complex noise backgrounds made fault diagnosis of hydraulic waterproof valves difficult.To address this issue,a fault diagnosis method of waterproof valves based on multi-sensor information fusion and a convolutional neural network-bidirectional gated recurrent unit-attention mechanism model was proposed.Firstly,considering that a single sensor's vibration signal might inadequately express fault characteristics,three sensors were employed to collect noisy vibration signals,and the necessary preprocessing was performed.Secondly,16 time-domain features,5 frequency-domain features and 3 time-frequency domain features of the signal were extracted.These features were fused using the entropy weight method to enhance their representativeness.Then the fused multi-dimensional feature set was input into the CNN-BIGRU-Attention model for feature recognition.Finally,the effectiveness of this method was validated through practical hydraulic waterproof valve fault diagnosis experiments.The research results indicate that features extracted with multiple sensors are more comprehensive.The fusion of information helps capture a more complete set of hidden data features,and significantly improves diagnostic accuracy.Comparing to other feature recognition methods,the fault diagnosis accuracy of hydraulic waterproof valves using the proposed method increased by 6.7%,4.6%,and 14.2%,reaching 96.86%,which proves the effectiveness of the method.This method provides a novel,efficient solution to a prevalent issue in construction engineering,combining advanced machine learning techniques with practical engineering applications.

关 键 词：液压传动系统 液压防水阀 多传感器 滑动时间窗 TEAGER能量算子 熵权法 卷积神经网络-双向门控循环单元-自注意力机制模型 

分 类 号：TH137.52[机械工程—机械制造及自动化] TH17[建筑科学—建筑技术科学] TU60