.png){kind=icon}
.png){kind=icon}
Return
Modeling vehicle dynamics with physics-informed deep operator network
.png){kind=icon}
PRE
AI
DOI：10.1080/00423114.2025.2526056.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In this paper, we present a novel data-driven approach for modelling vehicle dynamics that integrates the physics-informed deep operator network with the principles of linear time-invariant (LTI) state space modelling. This approach aims to achieve precise and robust data-driven vehicle dynamics by utilising vehicle physical models. These physical models have been validated experimentally and through rigorous theoretical derivations. The proposed approach employs a deep operator network to nonlinearly map the original state space into the LTI state space. Additionally, the deep operator network uses physical models as prior knowledge to establish the causality between data variables, thereby enhancing the interpretability of the model. Then, we use the linear relationship between vehicle states to develop a physics-informed loss function. This loss function effectively combines imperfect data with vehicle physics knowledge. The experimental results demonstrate that our proposed approach is enhanced by relatively precise prior knowledge. At the same time, data-driven methods can be used to complement the unmodeled vehicle dynamics. The approach is compared to existing methods and experimentally validated to demonstrate its effectiveness in improving the precision and generalisability of vehicle dynamics modelling.
Keywords:
Vehicle dynamics
physics-informed machine learning
state-space modelling
deep neural network
Journal
V
IF:
3.9
Papers:
3.1K
Citations:
8.9K