W artykule zostały przedstawione wyniki badań wpływu prędkości obrotowej i obciążenia na wartość momentu tarcia generowanego w węźle aparatu czterokulowego. Współpraca elementów pary ciernej odbywała się w szerokim zakresie obciążeń (500-6000 N) oraz prędkości obrotowych (100-2000 obr./min). Dokonano analizy uzyskanych wyników i podjęto próbę stworzenia trzystanowego klasyfikatora tarcia (stany smarowanie, zacieranie i zespawanie) z wykorzystaniem sztucznych sieci neuronowych. W celu uzyskania jak najlepszych wyników modelowania zastosowano różne rodzaje sztucznych sieci neuronowych, a także różne algorytmy uczące.
EN
Occurrence and synergy of many complicated friction phenomena, which have non-linear character, very often result in the lack of the possibility to forecast courses and the effects of tribological processes. Searching models of tribological quantities like wear, friction coefficient, the moment of friction, and temperature taking into account every process which proceed in friction pairs is one of the most important problem in present tribology. As far as boundary friction is concerned, the situation is even more difficult. We are not able to predict if and when the boundary layer is destroyed (which value of load and how much time is needed). In view of their properties, the artificial neural networks (ANN) could become very useful instruments. They let us to carry out some multidimensional analysis, define the influence of single parameters and what is important - the interaction between these parameters. Obtained results of the influence of load and rotational speed on the moment of friction and wear of a tribological pair are presented in the paper. Test were carried out at the range of a rotational speed about 100-2000 rpm and a load of about 500-6000 N. During the tests moment of friction, oil temperature and weather conditions were registered. After the tests the wear of the tribological pairs were measured. The analysis of results was elaborated and three-state friction classifier on base of artificial neural networks (ANN) was built. The different training algorithms were applied to obtain the best quality of built models. All researches showed that artificial neural networks are very useful as prediction models in tribological processes. The analysis proved that RBF networks are the most suitable for classification. The obtained model achieved quite good precision - RBF networks give higher than 90% quantity of correct classifications. Considered networks had a simple structure - only two inputs because teaching data was quite small (for example weather conditions were skipped). It was also proved that implementation of an additional state - "welding" deteriorated the classifiers quantity insensibly.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.