Fault diagnosis of mechanical drives under non-stationary conditions based on manifold learning of kernel mapping

作者：Liang Lin; Liu Dan; Wen Guangrui^*; Xu Guanghua; Zhang Yizhuo

来源：Journal of Vibroengineering, 2013, 15(1): 313-325.

• manifold learning
• fault detection
• kernel mapping
• non-stationary condition

For the detection of mechanical faults under the operating conditions of varying speeds and loads (such as wind turbines, excavators or helicopters, etc.), a new method for extracting the low-dimensional embedding of vibration data sets of mechanical drives under variable operation conditions is proposed. The hypothesis is that the space spanned by a set of vibration signals can be captured in a varying condition, to a close approximation, by a low-dimensional, nonlinear manifold. This paper presents a method to learn such a low-dimensional manifold from a given data set. The embedding manifold generated by vibration signals can be constructed from the feature set of parameters. Taking the variable operation condition into consideration, the kernel mapping is also introduced to improve the identification of submanifolds in terms of the projection distance. With the kernel mapping, the manifold coordinates can accurately capture the differences of the varying operation conditions. Experimental vibration signals obtained from normal and chipped tooth fault of gearbox in varying operation conditions are analyzed in this study. Results show that the proposed method is superior in identifying fault patterns and effective for gearbox condition monitoring.

• 出版日期2013-3
• 单位西安交通大学