Towards an Automated AI Based Simulation Framework in Aerospace Engineering

In the aerospace industry, there is an increasing inclination to replace costly and time-consuming prototyping with computer simulations.These simulations are characterized by their complexity and the necessity of not only a proper physical representation, but also the capacity to meet time-to-solution expectations.Numerous specialized software packages have been developed in recent years, incorporating a range of numerical techniques.However, the majority of these packages are designed to address specific problems such as structural mechanics or computational fluid dynamics problems.Real-world applications, however, are more involved and require the consideration of coupling approaches, which allow the combination of available tools to realize complex and interdisciplinary simulations, involving not only multi-scales, but also multi-physics.In this regard, supercomputing systems are indispensable, as they allow the realization of complex problems.The main question persist, how to enhance the efficiency of computations beyond the scaling with increased computational resources.Hererby, machine learning has the potential to play a pivotal role by not only reducing the computation but also allow for predicting the numerical solutions that do not necessitate extensive hours of computing time on supercomputing systems.This, in turn, has the effect of significantly reducing the time to solution, which is an important factor from the industrial perspective.However, it is important to ensure the reproducibility and reliability of the predicted results for critical decision-making processes.The processes involved in such an approach are complex and data-driven.The incorporation of this approach requires not only the software itself, but also data, batching, preprocessing, and other related components, to utilize the training and validation of the machine learning algorithm.In this context, provenance emerges as a crucial element in the documentation of the entire process, from the initial configuration of a simulation to the prediction of results and the identification of