A comparison of two symptom selection methods in vibration-based turbomachinery diagnostics

Complex diagnostic objects, e.g. critical rotating machines, usually generate a large number of diagnostic symptoms. A procedure is therefore required of selecting those most suitable from the point of view of technical condition evolution representation. A method based on the Singular Value Decomposition has been proposed for this purpose. An alternative is provided by an assessment of information content variation with time. Any symptom, treated as a random variable, may be assigned an information content measure that determines its 'predictability'. As the end of service life is approached, symptom value is to a growing extent dominated by deterministic (and hence predictable) lifetime consumption processes, which implies decreasing information content. The symptom with the fastest decrease of an information content measure with time should thus be judged the most representative one. Suitability of such approach has already been demonstrated. The aim of this paper is to compare results obtained with both methods.