RISK CDT - Reducing Uncertainty in Simulated Flight Tests by Increasing Flight Simulator Fidelity Through Machine Learning

PLEASE APPLY ONLINE TO THE SCHOOL OF ENGINEERING, PROVIDING THE PROJECT TITLE, NAME OF THE PRIMARY SUPERVISOR AND SELECT THE PROGRAMME CODE "EGPR" (PHD - SCHOOL OF ENGINEERING)

This is a project within the multi-disciplinary EPSRC and ESRC Centre for Doctoral Training (CDT) on Quantification and Management of Risk & Uncertainty in Complex Systems & Environments, within the Institute for Risk and Uncertainty. The studentship is granted for 4 years and includes, in the first year, a Master in Decision Making under Risk & Uncertainty. The project includes extensive collaboration with prime industry to build an optimal basis for employability.

Flight simulators are a highly integrated part of the aircraft life cycle. They are used in design and development phases, testing and qualification activities as well as in training and research. Their use continues to grow as the pilot training community is becoming increasingly reliant on simulation facilities. They can provide a proportion of a pilot’s initial and recurrent training, more quickly and more cheaply than in actual aircraft. They fulfil the increasing demand for new pilots to replace the current population of pilots as they reach retirement age. At the same time, in a military context, mission rehearsal (in air, land and sea contexts) is making increasing use of simulation.

A flight simulator consists of many components including visual, audio, and control systems as well as representative cockpit environments. At the heart of any flight simulation facility is the flight dynamics model of the aircraft being simulated. Techniques to design and develop such models are well known and documented. However, there is a requirement for the entire system to run in real-time and this can lead to simplifications having to be made, particularly for more complex aircraft such as rotorcraft. Often, the modelling techniques employed simply cannot capture all of the physical aspects of the real scenario. This can lead to significant differences between the response of the aircraft model and that of the real aircraft. This, at best, can be off-putting if it is noticeable to the pilots and, in the worst case scenario, can actually result in negative training being delivered to the crew flying the simulator.

Of course, standards related to the fidelity of such models already exist for flight simulators that are used for training purposes. However, it is still the case that the response of the aircraft model does not have to match that of the real aircraft exactly. There has been and continues to be significant effort at Liverpool on the topic of generating metrics to measure simulation fidelity. This project aims to complement that research theme through the use of machine learning techniques to develop a “grey-box” approach to flight simulation models for flight training/mission rehearsal and to assess any improvements that this brings to the simulation environment. In this approach, the physical models will be used to generate the aircraft response to the simulated pilot control inputs wherever possible. However, a machine-learned “black box”, that has been trained on real/truth flight data, will be used to ‘step-in’ when the simulated response deviates from the ideal. The project will investigate whether this can be achieved and, if so, how it can be implemented practically. Crucially, it will also establish the benefits and disadvantages of using such a technique. The project will adopt a handling-qualities approach to this problem, whereby clinical mission task elements will be defined for training purposes. Both objective and subjective measures will be generated from pilots who will fly these manoeuvres, using simulation models where the newly developed method is both active and inactive. In this way the project is attempting to both quantify and eliminate the uncertainty surrounding different pilot responses between simulation and flight by removing one of the key sources of error in the simulation chain i.e. the flight model response.