HPC as Enabler for the Virtualization of Aircraft Development
Guus Dekkers, CIO of EADS and Airbus
HPC as Enabler for the Virtualization of Aircraft Development It is a long journey from the initial idea towards a flying aircraft: The development of a new model from scratch requires billions of Euro’s worth of investments and consumes multiple years.
With only a handful of prototypes build towards the end of this process, it is clear that design flaws have to be avoided at any cost, as their correction late in the process requires substantial money and time and is in many cases even impossible to correct. This drives for a need to digitally model in a very early stage the different requirements the aircraft needs to comply to, in order to be capable find the optimal balance amongst those environmental, economical, safety, performance and operational parameters.
Beyond other means as design in context, numerical simulation is one of the most important means to realize this objective. To make this happen, the aircraft is modelled in an early phase representing both physical flight behaviour as well as the behaviour of the different aircrafts structures (ex. flaps) and systems (ex. powerplant). The design teams then apply different improvement plans and “what if” studies to this model, which allows for a quick analysis of the consequences of any modification or optimization suggested. As this is performed today with more and more detail and in shorter and shorter timeframes (example: A320 NEO), substantial HPC-power is needed.
There are at least two ways to perform such simulation, which ideally would happen on an almost “real time” basis. One consists of the definition of the overall aircraft model, which will ultimately translate into between 500 million to 1 billion nonlinear equations which have to be resolved during each reiteration in a time-accurate manner. This is a tremendous challenge requesting both substantial computing power and the further optimization of all calculation engines used, however, recent major numerical developments have shown good progress.
The other method exists of the modelling of hundreds of pre-computed solutions throughout the (flight) envelope which will form a proper basis for interpolation to represent the whole mission range. This works well for flight physics, but is not truly suited for an overall system-design.
Generally speaking, both ways have specific pros and cons depending on their specific used cases, both are still in the process of further evolvement and they clearly use HPC in a substantial different way.As Airbus has a vast interest in the further evolution of this field, we are actively pursuing advanced research in this area, in most cases in formalized programs like FUSIM. At large, we are cooperating with partners from recognized R&D institutes and universities active in this area with the ultimate objective to make aforementioned vision a reality.
Future Peta/ Exa-flops computer environments are an essential part of this approach, but only one element: At the same time the computing algorithms themselves need to face substantial evolvement in order to contribute to this vision as well. Recent progress in HPC, numerical algorithms and complex integrated simulation systems has shown a clear route to achieve this objective: accurately simulate flying aircraft with interaction of all relevant disciplines on board.