The present study describes a frequency spectrum classification method for fault detection of the LP gas pressure regulator using support vector machines (SVMs). Conventional diagnosis methods are inefficient because of problems such as significant noise and non-linearity of the detection mechanism. In order to address these problems, a machine learning method with a divergence-based kernel is introduced into spectrum classification. The authors use the normalised frequency spectrum directly as input with the divergence-based kernel. The proposed method is applied to the vibration spectrum classification of the rubber diaphragm in a pressure regulator. As a result, the classification performance using the divergence-based kernel is shown to be better than when using common kernels such as the Gaussian kernel or the polynomial kernel. The high classification performance is achieved by using an inexpensive sensor system and the machine learning method. The proposed method is widely applicable to other spectrum classification applications without limitation on the generality if the spectra are normalised.
ASJC Scopus subject areas
- Signal Processing
- Electrical and Electronic Engineering