Use of machine learning in grease evaluation
Competence Center Tribology
During the development of lubricating greases, numerous chemical-physical and mechanical-dynamic measurements and tests are carried out in order to determine the suitability for a planned application. Even today, the evaluation and interpretation of these measured values is still based on the experience of a few experts. Computer support for this process or even a computer simulation of real tribological systems running in boundary and mixed friction is not yet possible, apart from a few very scientific special cases. The diverse interactions between the surfaces, wear particles, lubricant and ambient medium in conjunction with the changes in the lubricant over time and non-linear system behavior cannot yet be simulated and predicted macroscopically. The reason for this is that the complex relationships cannot yet be adequately described mathematically/physically and extreme differences in size from atomic chemical processes to the macroscopic shape of the components (approx. ten powers of 10) have to be taken into account. The aim of the project is to demonstrate the potential of machine learning in lubricant development. As a concrete application example, the extensive data from the previous DGMK 820 project on thickener degeneration as a result of thermal, catalytic, oxidative and mechanical stress will be analyzed using various ML methods and algorithms. In this way, the boundary conditions, advantages and limitations of this innovative approach can be made tangible. The working hypothesis is that ML approaches make it possible to evaluate measurement data in lubricant evaluation more objectively, quickly and reproducibly and to recognize complex relationships that remain hidden to humans due to their limitation to 3 dimensions.
The IGF application for assessment was approved. The application for funding was submitted to the BMWK.