top of page


Mr. John Smith

Job title



Co-simulation techniques are nowadays widespread in the model based development field. These techniques allow simulation engineers to capture complex phenomena by leveraging on the strength of any individual simulation model. Co-simulation models tend to be very accurate, but rather complex and time consuming to evaluate. On the other hand, some applications such as HILS or driving simulators require light models that can be executed in real time. This paper considers the feasibility of the usage of Machine Learning based Reduced order modelling approaches for reducing the complexity of a co-simulation model. The co-simulation model evaluated in this paper includes two multi body simulation sub-models coupled with a look up table based logic sub-model. Specifically, this paper presents an alternative technique in which the logic model is replaced with a machine learning model, with the aim of improving real time performances of the whole co-simulation model, while maintaining an accuracy sufficient for the application case at hand. Different types of machine learning models have been investigated, including recurrent neural networks, random forest based models and Gaussian Process regression models. Moreover, different training data gathering approaches for the creation of the machine learning training dataset have been tested. Lastly, the various evaluation methods used to assess the machine learning model quality and their impact on the causality links of the model is discussed. As results of this investigation, this paper highlights the limitations of using the traditional evaluation methods for machine learning models in the scenario where the target application case is a co-simulation between simulation models and machine-learning ones. A novel approach for Machine Learning model training and testing has been proposed, which resulted in significant accuracy improvements in the selected study case. In conclusion, a multi-body and Machine Learning coupling scheme has been deployed on the driving simulation achieving real time performances and satisfactory accuracy.

Mr. Roberto Bossio, Engineer, Toyota

Machine Learning based Reduced Order Models for Real Time application

FWC2023-DGT-018 • FISITA World Congress 2023 • Digitalisation


Sign up or login to the ICC to download this item and access the entire FISITA library.

Upgrade your ICC subscription to access all Library items.

Congratulations! Your ICC subscription gives you complete access to the FISITA Library.


Retrieving info...

Available for purchase on the FISITA Store


bottom of page