Effects of input features in support vector machine based rolling element bearing fault detection and trending

Abstract eng:
Rolling Element Bearings consist one of the most widely used industrial machine elements, being the interface between the stationary and the rotating part of the machine. Due to their importance a plethora of monitoring methods and fault diagnosis procedures have been developed, in order to reduce maintenance costs, improve productivity, and prevent malfunctions and failures during operation which could lead to the downtime of the machine. Towards this direction, among different automatic diagnostic methods, the Support Vector Machine (SVM) method has been shown to present a number of advantages. Support Vector Machine is a relatively new computational learning method based on Statistical Learning Theory and combines fundamental concepts and principles related to learning, well-defined formulation and self-consistent mathematical theory. The key aspects about the use of SVMs as a rolling element bearing health monitoring tool are the lack of actual experimental data, the optimal selection of the type and the number of input features, and the correct selection of the kernel function and its corresponding parameters. A large number of input features have been proposed, being divided in two big categories: A) Traditional signal statistical features in the time domain, such as mean value, rms value, variance, skewness, kurtosis etc, B) Frequency domain based indices, such as energy values obtained at characteristic frequency bands of the measured and the demodulated signals. In this paper, the structure and the performance of a Support Vector Machine based approach for rolling element bearing fault diagnosis is presented. The main advantage of this method is that the training of the SVM is based on a model describing the dynamic behavior of a defective rolling element bearing, enabling thus the direct application of the SVM to industrial measurements of defective bearings, without the need of training the SVM with experimental data of a defective bearing. Moreover, the different types of input features (i.e. traditional statistical time domain indices versus frequency domain indices) are evaluated and compared, indicating a superiority of the frequency domain features.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket