摘要
针对现有转子故障诊断方法缺乏多尺度特征提取能力以及强噪声导致故障分类准确率较低的问题,提出一种基于格拉姆角差场(Gramian Angular Difference Field,GADF)、Inception模块与锐度感知最小化(Sharpness Awareness Minimization,SAM)的故障诊断方法。首先,对故障信号进行GADF变换,并使用多传感器信息融合策略将由不同传感器信号变换得到的图像进行水平拼接;接着,将拼接后的图像输入到Inception-SAM模型中进行分类识别,其中,Inception模块能增强神经网络对于多尺度特征的提取能力,SAM能增强模型的泛化性能。实验结果表明,所提方法在转子故障诊断中的分类准确率能够达到99.64%,而且相比其他图片编码方法和神经网络模型,该方法具有更高的故障分类准确率;同时,抗噪性能测试证明该方法在高噪声条件下仍具有较高的准确率。
To address the problems that the existing rotor fault diagnosis methods lack the ability of multi-scale feature extraction and the strong noise leads to the low accuracy of fault classification,a fault diagnosis method based on GADF-In-ception-SAM was proposed.Firstly,the fault signal was transformed by GADF,and the images obtained by transforming dif-ferent sensor signals were horizontally spliced using a multi-sensor information fusion strategy.Then,the spliced images were input into the Inception-SAM model for classification and identification,in which the Inception module enhances the neural network ability for extracting multi-scale features,and the SAM enhances the generalization performance of the mod-el.The experimental results show that the proposed method can achieve 99.64%classification accuracy in rotor fault diagno-sis,and it has the highest fault classification accuracy compared with other image coding methods and neural network mod-els.Meanwhile,the anti-noise performance test proved that this method still has high accuracy in the high noise condition.
作者
伍济钢
张源
曾嘉
WU Jigang;ZHANG Yuan;ZENG Jia(School of Mechanical Engineering,Hunan University of Science and Technology,Xiangtan 411201,Hunan,China)
出处
《噪声与振动控制》
北大核心
2025年第3期98-104,221,共8页
Noise and Vibration Control
基金
国家自然科学基金资助项目(51775181)。
关键词
故障诊断
格拉姆角差场
多传感器信息融合
锐度感知最小化
转子
fault diagnosis
Gramian angular difference field
multi sensor information fusion
sharpness awareness minimization
rotor
