Zastosowanie analizy topologicznej obrazów termowizyjnych do oceny zmian stanu technicznego maszyny wirnikowej

Artykuł dotyczy oceny przydatności analizy topologicznej obrazów termowizyjnych do ciągłego diagnozowania maszyn i urządzeń. W trakcie badań rozpatrywano zbiór 81 cech topologicznych, z którego stosując dyskryminacyjną analizę Fishera, wyselekcjonowano podzbiór 10 cech relewantnych. Analizie poddano sekwencję obrazów termowizyjnych maszyny wirnikowej, zarejestrowanych w trakcie czynnego eksperymentu diagnostycznego. Wyznaczone cechy pozwoliły na wyznaczenie sygnałów diagnostycznych, których ocena pozwala na identyfikację stanu technicznego maszyny.

In the paper results of investigations aimed at verification of usefulness of applying the thermogram topological analysis to continuous diagnostics of machines and devices are presented. The analysis was applied to sequences of 840 thermograms of the rotating machinery model obtained from active diagnostic experiments performed in a laboratory of the Department of Fundamentals of Machinery Design. During the experiment six technical conditions of the object were simulated. At least three of them were very similar to each other and were simulated in order to check the possibility of detecting weak changes of the condition. The thermogram was analyzed within two Regions of Interest (ROI) representing machine bearing housings of the same type. The set of 81 features was calculated for each ROI. Consideration of two ROIs representing the same type of the bearing housing was expected to be beneficial for verification of different image features and condition classifiers. Due to the huge number of thermogram features it was necessary to select the subsets of relevant features characterized by the optimal performance. The classical feature selection method based on Fisher discriminat analysis was used to limit the set to 10 relevant features. The estimated features allow generating diagnostic signals enabling classification of the machine technical conditions. An artificial neural network was used as a condition classifier. Very promising results of the classification confirmed the usefulness of selected topological features to estimation and recognition of the machine technical condition.