Work

Improve Dynamics of Autonomous Vehicles During Loading and Unloading of Bulk Material - Data-Driven Reduced Order Models Derived from DEM Simulations Based on Algoryx Dynamics

Year: 2026
Type: dissertation
Abstract: The rapid development of automated driving is transforming confined-area transport systems, and it requires extensive simulations to ensure safety of autonomous heavy-duty vehicles. However, modelling bulk materials such as rocks and sand using high-fidelity Discrete Element Methods (DEM) is computationally heavy and time-consuming, making it impractical for real-time applications. To address this problem, this thesis develops a machine learning-based reduced order model (ROM) in Altair RomAI to predict 3 forces and 3 torques on the pivot hinge during loading and unloading scenarios. Training datasets with different particle masses, tipping angles, and floor inclinations were generated via AGX Dynamics simulations. To optimize the network architecture, a hyperparameter tuning process was conducted on a subset of the whole training set. The trained ROM shows great fitting and robust generalization on unseen validation scenarios. The model successfully captured the physical trends and main unloading dynamics. Additionally, the ROM runs in several seconds compared to the minutes required by AGX Dynamics simulations, significantly reducing time cost. Finally, the model was exported as a Functional Mock-Up Unit (FMU) and integrated into Volvo’s downstream vehicle dynamics framework. Serving as a real-time virtual sensor, the ROM accurately outputs the forces and torques into the complete vehicle model.
Author: Du, Shen
Institution: Aalto University
Language: English
Open Access status: green
OpenAlex ID: W7203956794
Cites: 22
Cited by:
Related to: 10