Churning loss modeling for shroud gears driven by physics-data hybrids

Abstract

The complexity of multiphase physical mechanisms limits the development of precise mathematical models for churning losses in shrouded gears. This research aims to characterize the impact of shrouds on gear churning losses through a combined data- and physics-driven approach and to develop a mathematical model suitable for complex mechanical systems. Initially, a physical model for churning losses in bare gears was established by simplifying the Navier-Stokes equations, providing a foundation for modeling under shrouded conditions. Subsequently, a mathematical model, the churning loss coefficients model (CLCM), was developed using least squares fitting. This model incorporates seven coefficients to be determined. After numerous tries, the presented CLCM adapts to multiple shroud configurations considering the evolution of the gear speeds. In addition, a data-driven optimization framework for the CLCM was designed. It integrates a backpropagation neural network, optimized by particle swarm optimization, with third-generation non-dominated sorting genetic algorithms. This framework, which integrates the capabilities of accurate description of multiphase physical processes and comprehensive multi-objective search, enables the comprehensive optimization of the primary coefficients in the CLCM across multiple speeds. Finally, the generalization capability of the optimized CLCM was validated on a complex planetary gear system, demonstrating a prediction accuracy exceeding 93 %. These findings provide valuable insights for the development of more efficient gearboxes.