期刊
INTERNATIONAL JOURNAL OF ADVANCED MANUFACTURING TECHNOLOGY
卷 106, 期 7-8, 页码 3063-3070出版社
SPRINGER LONDON LTD
DOI: 10.1007/s00170-019-04858-w
关键词
Artificial intelligence; Ball bearings; Bootstrap resampling; Complementary ensemble empirical mode decomposition; Classifications; Fault diagnosis; Fault size estimation; Kernel support vectormachines
The successful assessment of the health condition in rolling element bearings hinges on the early fault detection of fault of bearing elements. Early research has demonstrated that interactions of various components within the mechanical system and added noise from data acquisition system result in the unclear trend of the vibration signal from bearings. The unclear trend could lead to inaccurate assessment of the machine condition, which could result in profit loss and catastrophic failure. Therefore, obtaining the machine condition under heavy noises in the early degradation stage is critical to ensure smooth operation and maximize productivity. In the past, various signal processing techniques are implemented in bearing diagnosis to improve the signal to noise ratio and monitor frequency bands that are associated with bearing fault. The diagnostic models generally involve parameter tuning and feature selections. The tuning of the parameters of the diagnostic model could be tedious and require understanding of the characteristics of the acquired signal. Besides the generally selected signal features such as root mean square, kurtosis, and skewness, the selection of other features from signals does not provide much insights into the degradation of bearings. Features are mixed in a way that a few of the features could have negative impact on the assessment of degradation. To address the abovementioned issue, we present an innovative diagnosis model using the complementary ensemble empirical mode decomposition (CEEMD) with kernel support vector machines (kernel SVM) to evaluate the health condition of bearings in terms of defect severity. To the best of our knowledge, the combination of the two methods has not been implemented in documented research work. The novelty of this work is the optimization of the CEEMD parameter using the bootstrap resampling method while combining the kernel SVM to characterize the fault size of the bearings. The acquired signal is processed using the complementary ensemble empirical mode decomposition method to remove undesirable noises and extract the signal containing the fault signature. The signals are then examined, segmented, and classified based on the degradation stage of bearings. The kernel SVM is implemented to learn the correct classification and predict the corresponding bearing degradation stage based on the simple root mean square value instead of using various features. The model has a prediction accuracy of about 98% and is easy to implement because of less parameters involved. In comparison with the traditional SVM and kernel SVM, the proposed model significantly improved the accuracy of prediction. Because of its high prediction accuracy and easy implementation, the proposed model can be implemented into machine condition monitoring on CNC production machines and various types of rotating machineries.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据