As a system’s response, wear behavior is influenced by material properties and additional factors such as lubrication conditions. Modeling and predicting a material's wear behavior with traditional numerical methods is difficult due to this complexity. Recent developments in machine learning offer the possibility of solving higher-order, nonlinear problems.

In our project, the complex pattern of the wear test data will be used for the categorization of wear mechanisms using data-driven methods. Wear tests are carried out with different materials for data acquisition. Several statistical analyses are then performed on the recorded normal force, friction force, and friction coefficient to extract appropriate features. A clustering algorithm is implemented to categorize the wear behavior into several clusters. In the following, a similarity equation is formulated based on the statistical distribution of the features. Finally, the wear behavior of new materials is predicted as a weighted combination of the categories.

The statistical analysis reveals the correlations between several features and the varying wear behavior of Ti-, Al-, and Fe-based alloys. For instance, the distribution of friction force and normal force differs depending on the wear behavior, as illustrated in the figure below. These simple statistical features can contribute to roughly distinguish the wear behavior. Future work will include additional information from the tribometer such as oscillations and dynamics, as well as physical properties of the materials and lubricant information. This will enable a more detailed categorization of the wear behavior.