Method of classification of global machine conditions based on spectral features of infrared images and classifiers fusion

This paper describes an original method of global machine condition assessment for infrared condition monitoring and diagnostics systems. This method integrates two approaches: the first is processing and analysis of infrared images in the frequency domain by the use of 2D Fourier transform and a set of F-image features, the second uses fusion of classification results obtained independently for F-image features. To find the best condition assessment solution, the two different types of classifiers, k-nearest neighbours and support vector machine, as well as data fusion method based on Dezert–Smarandache theory have been investigated. This method has been verified using infrared images recorded during experiments performed on the laboratory model of rotating machinery. The results obtained during the research confirm that the method could be successfully used for the identification of operational conditions that are difficult to be recognised.