Abstract
Physics-based models are widely used to analyze transient responses of nonlinear dynamic systems across engineering applications, including vessel dynamics, robotics, and energy systems. These models preserve physical consistency and provide reliable dynamic behavior, but their repeated numerical evaluation over varying operating conditions can become computationally expensive due to time integration. This limitation is particularly important in applications requiring rapid short-horizon predictions and repeated queries, such as real-time control, autonomous decision-making, and digital-twin updates. Developing efficient machine-learning-based surrogates for such settings is therefore of significant practical interest.This study proposes a machine-learning-based framework for learning nonlinear transient responses directly from simulation data generated by a validated system-based physical model. Rather than replacing the underlying physics, the approach represents its parametric input-output behavior through a time-conditioned surrogate formulation that enables direct finite-horizon response emulation. Tree-based ensemble methods are employed to model the relationship between system inputs, time, and the resulting transient states, with ship maneuvering dynamics used as a representative class of nonlinear transient-response problems.The results show that tree-based ensemble surrogates can accurately reproduce short-horizon transient responses across varying operating conditions while providing substantial computational speedup relative to the underlying physics-based model. The models preserve both the dynamic-state evolution and the associated kinematic behavior, enabling reliable trajectory reconstruction. In addition, the framework remains effective under significantly reduced training data, indicating strong data efficiency. Overall, the findings identify tree-based ensemble methods as a practical, accurate, and computationally efficient solution family for finite-horizon emulation of transient responses in nonlinear dynamic engineering systems.