| dc.description.abstract | Abstract. Drag torque generated in disengaged wet clutches constitutes a critical factor causing power loss, efficiency reduction, and reliability issues in transmission systems. To achieve accurate prediction of drag torque and associated power losses, this study proposes a hybrid modeling approach integrating tribological mechanisms with machine learning. First, based on tribological principles, drag torque calculation models were established for low-speed zones and high-speed zones; simultaneously, a data-driven model was constructed using a multilayer perceptron (MLP). Subsequently, inverse variance weighting was employed to fuse predictions from these mechanistic and MLP models through weighted integration, with the combined model demonstrating significantly superior accuracy over any single model. To further enhance performance, an exponential triangular optimization (ETO) algorithm was introduced to optimize hyperparameters of the hybrid model, and its reliability was validated through systematic experiments covering the wide speed range (0–5000 rpm) and multiple parameter combinations including oil temperature, friction pair clearance, and oil supply flowrate. Using the established model, a systematic investigation was conducted on the influence patterns of oil temperature, friction pair clearance, and oil supply flowrate on drag torque and power loss. Results demonstrate that increasing oil temperature or enlarging friction pair clearance reduces the critical rotational speed corresponding to the drag torque and power loss recovery point, thereby diminishing drag torque and power loss; conversely, increasing oil supply flowrate elevates this critical speed and exacerbates drag torque and power loss. | |
