Machine Learning-Based Prediction of Dynamic Braking Torques in Dry-Running Friction Systems - A Data-Driven Approach to Optimize Industrial Brake Systems

Resource inefficiencies in safety-critical systems—such as industrial brakes in passenger elevators—often result from conservative design approaches caused by uncertainties in predicting frictional behavior. Established industrial methods, including flywheel tests and partial friction lining tests, frequently fail to capture the full complexity of frictional interactions under realistic operating conditions influenced by temperature variations, surface pressure, sliding speeds, and surface conditions. Moreover, conventional numerical methods, such as finite element analyses, are only partially capable of modeling the complex and highly non-linear effects in frictional contact. This paper introduces a data-driven approach leveraging machine learning techniques to predict the dynamic braking torque curves and variability of the coefficient of friction under varying operating conditions. The methodology is based on a modified CRISP-DM process, augmented by statistical analysis, dimensionality reduction, and clustering techniques. Furthermore, a hybrid modeling approach systematically integrates physical-tribological prior knowledge via simulation results (FEM, physical-technical simulations in Simulink) or analytical formulations, significantly enhancing prediction accuracy, especially in data-sparse operational regions. Additionally, the paper introduces a novel clustering-based approach that selects tailored ML models for specific braking scenarios, further improving predictive performance. Particular emphasis is placed on transfer learning methods to generalize acquired model knowledge efficiently to different industrial braking systems, significantly reducing experimental effort required for new system development and validation. The developed models also provide uncertainty quantifications through prediction intervals, allowing reliable assessments of braking torque predictions. Consequently, this integrative framework significantly reduces oversizing and resource consumption, actively contributing to sustainability through reduced material and energy use in brake system development.