Classifying wear characteristics in lubricated sliding wear based on time series sensor signals using artificial intelligence

点击翻译按钮获取中文摘要

Friction and wear cause high economical losses in terms of energy and resources still today. Estimated numbers for losses due to friction and wear range at ≈23% of global energy consumption. While friction is assumed to cause significantly higher direct costs than wear, the adverse effects of wear are more difficult to describe, including amongst others aspects of environmental pollution or equipment downtime.Friction occurring in sliding surfaces usually reduces the efficiency of the technical system. Frictional energy is partially converted into thermal energy and enables the formation of wear particles. In addition, there are interactions with the dynamic behavior (stiffness, inertia) of the technical system in which the friction occurs. While a strong influence of material reactions - including wear - on the friction coefficient is clearly observed, no fundamental and general correlation between friction signal and wear mechanisms has been found, due to the high number of influencing parameters acting in different tribosystems. Still, in order to apply measures to mitigate wear, it is important to understand the wear mechanisms. Using easy to measure quantities such as friction force, temperature or vibrations of a machine to identify wear mechanisms taking place in a tribosystem - in the future also operando - is highly desirable.Novel, data-driven methods are a promising opportunity to address the intrinsic complexity of tribological problems. These include artificial intelligence (AI) and machine learning methods.In this project it will be investigated how reliably an AI can classify characteristic combinations of acting wear mechanisms and the resulting wear volumes, based on normal and friction forces, temperature evolution or high-frequency vibrations occurring in two different tribometers.To achieve these aims, high numbers of tribological tests are carried out in a controlled manner, yielding up to 5 classes of wear characteristics and wear volumes to train the AI. Their use on unseen data will show the potential of AI to classify wear characteristics and in the future possibly single wear mechanisms observed in sliding wear of metals on a fundamental level.