强背景噪声下故障特征分析与动力学时频联合校验方法研究  

Research on Fault Characterization Analysis and Kinetic Time-Frequency Joint Calibration Under Strong Background Noise

作　　者：赵乾柏 张春良[1] 岳夏[1] 龙尚斌 ZHAO Qianbai;ZHANG Chunliang;YUE Xia;LONG Shangbin(School of Mechanical and Electrical Engineering,Guangzhou University,Guangzhou Guangdong 510006,China)

机构地区：[1]广州大学机械与电气工程学院,广东广州510006

出　　处：《机床与液压》2025年第3期210-219,共10页Machine Tool & Hydraulics

基　　金：国家自然科学基金面上项目(52275097);广东华中科技大学工业技术研究院/广东省制造装备数字化重点实验室开放基金(2023B1212060012)。

摘　　要：针对现有采用神经网络进行故障诊断难以准确解释、精确寻找故障点的问题,提出一种基于动力学时频联合校验的强背景噪声下故障诊断方法。首先,利用连续小波变换对含噪声的振动信号进行时频转换,获取二维时频图;计算齿轮和轴承的故障周期,得到一系列的标准故障组合框;然后采用多个网络模型提取故障特征,并用热力图法对每个最佳权重网络进行结合,得到含注意力权重热力图的故障时频图,并在时频图对故障点进行时频特征提取,得到一系列的故障框;随后计算每个故障框的置信度并且取平均值,采用余弦相似度对捕捉故障框序列和标准故障组合框的位置进行评价,取置信度和位置的综合评价作为故障诊断的结果。最后分别利用凯斯西储大学和Spectra Quest行星齿轮箱数据进行实验验证。结果表明,在强噪声情况下,所提方法具有更高的识别效果和抗噪性,并可对故障过程进行准确解释。Aiming at the existing problem that it is difficult to accurately interpret and accurately find the fault point by using neural network for fault diagnosis,a diagnostic method based on kinetic time-frequency joint calibration of fault characteristics under strong background noise was proposed.The time-frequency conversion of the vibration signal contained noise was carried out using continuous wavelet transform to obtain a 2D time-frequency diagram.The fault cycles of gears and bearings were calculated to obtain a series of combining frames;then multiple network models were used to extract the fault features,and each best-weighted network was combined by the heat map method to obtain the fault time-frequency map containing the attention-weighted heat map,and the time-frequency features of the fault points were extracted in the time-frequency map to obtain a series of fault frames.Subsequently,the confidence level of each fault frame was calculated and averaged,and the cosine similarity was used to evaluate the fault frame and the combined frame in terms of location,and the combined evaluation of confidence level and location was taken as the fault diagnosis result.Finally,validation experiments were conducted using Case Western Reserve University data and Spectra Quest planetary gearbox data,respectively.The results show that in the case of strong noise,the proposed method has a higher recognition effect and noise immunity,and can accurately explain the fault process.

关 键 词：故障解释 故障特征分析 时频联合校验 

分 类 号：TH165.3[机械工程—机械制造及自动化]