【作者】 吴国文； 田杨阳； 毛文涛；

【Author】 WU Guowen;TIAN Yangyang;Mao Wentao;College of Computer Science and Technology, Donghua University;

【机构】 东华大学计算机科学与技术；

【摘要】 随着物联网等先进传感技术的快速发展,对轴承实时故障诊断的需求与日俱增。为提高轴承在线故障诊断的实时性和数值稳定性,本文提出一种基于特征融合的在线诊断方法。该方法利用核主成分分析算法对经验模态分解和小波包分解提取到的组合特征进行特征融合,并采用在线极限学习机对得到的融合特征构建在线诊断模型。整个过程分为离线和在线两个阶段:离线阶段,对提取到的不同故障状态下轴承数据的经验模态和小波包特征进行组合,并通过核主成分分析得到离线阶段的融合特征向量,构建初始在线极限学习机模型;在线阶段,使用两种算法对贯序到达的原始信号提取特征,并根据离线阶段获得的特征权重对其直接降维,从而动态更新在线极限学习机模型。在IMS轴承数据集上的实验结果表明,该方法能充分利用不同信息的互补特性,在不明显增加反应时间的情况下,有效提高了在线诊断精度与稳定性。更多还原

【Abstract】 With the rapid development of the Internet of things and other advanced sensing technologies, the demand for real-time fault diagnosis of bearings is increasing a lot. In order to improve the real-time and numerical stability of on-line fault diagnosis of bearings, an online diagnosis method based on feature fusion is proposed in this paper. The method uses Kernel Principal Component Analysis(KPCA) to fuse feature combinations extracted by Empirical Mode Decomposition(EMD) and Wavelet Packet Transform(WPT), and uses the online extreme learning machine to construct the online diagnosis model. The whole process can be divided into offline stage and online stage: in off-line stage, combining EMD and WPT features under different fault status, the kernel feature vectors are obtained by KPCA, and then the initial online extreme learning machine model is constructed; In online stage, EMD and WPT are firstly used to extract features for sequentially arrived raw signals, and according to the feature’s weights which are obtained in offline stage, the new features are reduced directly and the online extreme learning machine model is updated adaptively. The experimental results on the IMS bearing data set show that the method can make full use of the complementary characteristics of different information, and effectively improve the accuracy and stability of online diagnosis without increasing the reaction time.更多还原