Vehicle Simulation Laboratory

Research

Current Projects

Identification of Full-Stall Flight Models Motion Cueing for Full-Stall Recovery Fidelity Effects on Training Effectiveness MPC Motion for Driving Simulator Advanced Identification Methods for Flight Models

Identification of Full-Stall Flight Models

Identification of Full-Stall Flight Models from Certification Test Data

A large number of recent crashes have involved aerodynamic stall followed by inappropriate pilot control inputs. Therefore, most aviation governing bodies have extended pilot training to include full-stall recovery training in flight simulators. Unfortunately, due to the complicated flow conditions during stall, it is extremely expensive and difficult to develop accurate flight models, and therefore upcoming regulations only require a representative flight model in the stall regime. The VSL group, supported by the US FAA, has been developing methods for identification of these full-stall models from existing flight certification data. This is a very challenging problem as the certification flight test data is not well suited for automated identification. We have developed a new two-step multi-maneuver, robust MAP estimation algorithm that is well suited for this difficult identification problem.

Motion Cueing for Full-Stall Recovery

Motion Cueing for Full-Stall Recovery Training

Training simulators use motion systems to provide the pilots with motion cues that help them close the control loop around the simulated aircraft. The motion drive algorithm (MDA) filters the calculated aircraft motions so they fit within the motion system envelope while still providing good motion cues. The severe motions of full-stall recoveries require changes to the MDAs. We recently developed a new adaptive MDA that was shown to improve pilot performance during stall recovery.

Adaptive algorithm

Fidelity Effects on Training Effectiveness

Flight Model Fidelity Effects on Training Effectiveness

graph measuring task and recov score with nominal, severe and mild stall models

Due to the difficulty in modeling aerodynamic stall, current regulations only require representative flight models in the post-stall regime. Unfortunately the level of fidelity required for effective training is difficult to determine. Thus, the VSL recently ran a quasi-transfer of training study in their flight simulator to examine the impact of post-stall model fidelity on transfer of training. The experiment found no operationally significant differences in training between a nominal stall model (NM), a severe stall model (SM), and a mild stall model (MM).

MPC Motion for Driving Simulator

MPC Motion Drive Algorithm for Driving Simulator

Diagram for MPC Motion Drive Algorithm for Driving Simulator

While it has been shown that with proper tuning and parameter switching classical motion drive algorithms can achieve very high fidelity, the tuning process requires a highly skilled motion expert. Model Predictive Control has the potential to produce high-fidelity motion with substantially less tuning if the appropriate cost function can be found. We have recently developed and tested a new MPC with a cost function that includes an improved human motion perception model. For most driving maneuvers the new MPC algorithm was rated more highly than a well-tuned classical algorithm.

MPC Motion Drive Algorithm

Advanced Identification Methods for Flight Models

Helicopter

The VSL is working with CAE on advanced system identification algorithms for fixed and rotary wing aircraft flight models for use in training simulators. Flight models are often developed by selecting the parameters of a model such that its behaviour matches recorded flight test data from the real aircraft. Outfitting the test aircraft with a vast array of instrumentation and flying a large set of test maneuvers is extremely time consuming and expensive. In addition, tuning the parameters of the flight model such that its behaviour matches the real aircraft, to within required tolerances, is also extremely time-consuming process that requires significant involvement of expensive subject matter experts. This project aims to develop and apply new advanced system identification algorithms that will automatically estimate the parameters in a flight model such that its behaviour matches the recorded flight test data in an optimal sense. In addition, the same system identification algorithms will be used to determine the minimum airdata instrumentation requirements for the flight test aircraft that will result in a flight model with sufficient fidelity to comply with the appropriate regulations.