RSS feed LinkedIn Watch us on Youtube

Activity title

Enhanced Computational Performance and Stability & Control Prediction for NATO Military Vehicles

Activity Reference



Applied Vehicle Technology

Security Classification




Activity type


Start date


End date



control system design, data fusion, machine learning, multifidelity, reduced order models, stability control prediction, surrogate models


The prediction of the Performance and Stability & Control or Maneuvering characteristics (S&C) is essential for the design and performance assessment of modern military vehicles in the air and maritime domains. Previous activities focused on the validation of Computational Fluid Dynamics (CFD) methods and their applicability during early and advanced design stages and showed consistent results compared to experiments in many operational conditions. However, much effort is needed to provide comprehensive high-fidelity CFD-based data sets, both for capturing the nonlinear flow regime and for evaluating dynamical motions of the vehicle. Thus, current numerical capabilities are inadequate for efficiently predicting the extensive amount of data. Here, Reduced Order Modelling (ROM) approaches seem to be capable in predicting these complex behaviors at affordable costs by using a low-dimensional approximation of the response based on a few highly accurate CFD simulations.


The objectives are: to investigate and compare various numerical methods to create accurate performance and S&C data sets at feasible cost, to investigate methods for data fusion of numerical and experimental sets, to determine how to construct low-computational cost surrogate models of vehicle performance, and to investigate the applicability of aforementioned models within control system design.


(1) The efficient creation of S&C data sets based on highly accurate numerical CFD methods is of great interest. Various Reduced Order Models (ROMs) developed from these data need to be evaluated in their capability to accomplish this task, e.g. (a) Indicial response functions, (b) single/-multi fidelity surrogate models, (c) Proper Orthogonal Decomposition (POD) and/or Isomap-based ROMs, (d) neural networks. (2) Comparison of these models with established linear and/or nonlinear system identification models. (3) Assessment of reasonable training signals for aforementioned methods. (4) Data fusion of numerical and experimental data sources to enhance ROMs. (5) Application of potential beneficial models during control system design.

Contact Panel Office