SOW.pdf

Statement of Work

Government’s Minimum Need

Summary:

Statement of Work for MIT for Adaptive Envelope Protection Research Collaboration with TTT AEP NASA PI Shaun McWherter

The objective of this NASA funded effort is to allocate TTT (Transformative Technologies and Tools) project funding to support a one-year research collaboration between Dr. Chuchu Fan at the Reliable Autonomous Systems Lab at MIT (REALM) and Shaun McWherter, the NASA TTT Principal Investigator for the Adaptive Envelope Protection project. The objective of the effort will be to identify, evaluate, compare the best candidate mathematical methods and computational frameworks for implementing data-driven Adaptive Envelope Protection methods as a means for providing run-time safety assurance for an eVTOL (electric Vertical Take Off and Landing) aircraft during the flight test process which could involve model validation, system identification and control law adaption to expand the flight envelope.

The problem the Adaptive Envelope Protection research project aims to address for the NASA TTT project:

Many non-traditional eVTOL aircraft design are emerging that utilize multiple distributed electric motors and vehicle configurations that change during transitions between vertical and forward flight. These types of designs can present significant challenges to creating reliable aero-dynamic models and model-based control design.

Unreliable aerodynamic models dramatically increase the required engineering analysis during the flight test process and can drive overly conservative control law design that can limit vehicle’s maneuvering performance.

Fly-by-wire flight control systems integrated into eVTOL aircraft provide new possibilities such as varying the control allocation between rotors and control surfaces for different performance objectives as well as responding to failures, introducing complex control excitations in a precise and consistent manner during flight test, and automating envelope expansion, data acquisition and model and control validation activities.

One of the objectives of the TTT RAM (Rapid Aerodynamic Modeling) MCAAD (Modeling and Control for Agile Aircraft Development) project, which the Adaptive Envelope Protection project supports, is to enable rapid aero modeling by integrating online system identification and learning control methods to increase the tempo and efficiency of the envelope expansion, model validation and control design process.

The Adaptive Envelope Protection research aims to develop an independent runtime safety assurance capability to mitigate the loss-of-control risk that is agnostic to the system-ID and learning control methods employed during flight test. This would allow an increased in the tempo of the envelope expansion process by providing independent real-time safety guarantees.

The Adaptive Envelope Protection research goal is to develop and demonstrate mathematical methods for determining formally safe local flight envelopes and control policies online with minimal reliance on a-priori model information and maximizing the online utilization of flight data that are broadly applicable to different eVTOL vehicle designs.

If this is not technically completely feasible, the alternative goal is to develop and implement an adaptive envelope protection method based on the probabilistic quantification of safety relative to a model confidence quantification in critical areas of the flight envelope for eVTOL vehicles, such as hover, forward flight, and transitions in between that include changes in the overall vehicle configuration (e.g. tiltwing).

Requirement(s):

In support of the above objectives and research goals, MIT will perform research toward the implementation, evaluation, and comparison of adaptive envelope protection methods in a software simulation of a tiltwing eVTOL aircraft. Although not in the scope of this effort, the implementation and demonstration of the methods on an actual tiltwing vehicle is the goal. This goal should thus drive all considerations and frame the approach in developing, implementing, and evaluating the methods.

For this effort the definition of safety will include provably ensuring feasible control policies that can intervene and stabilize the aircraft to a constant level attitude constant velocity trim condition followed by a controllable descent and soft landing at any time, as well as prevent the vehicle from losing this safety assurance due to any maneuver, control excitation or external disturbance, especially during the flight-mode transition between hover to forward flight and back to hover.

The research effort will develop an independent runtime safety assurance capability to mitigate the loss-of-control risk that is independent of various system-ID and learning control methods employed during flight test. Thereby enabling a more efficient envelope expansion process by providing real-time safety guarantees for experimental learning controllers.

The research effort will aim to develop and demonstrate mathematical methods for computing online, provably safe local flight envelopes and control policies that require minimal reliance on a-priori models and maximize the real-time utilization of flight data. If this is not technically completely feasible, the alternative goal is to develop and implement an adaptive envelope protection method based on a probabilistic quantification of safety relative to a quantitative model confidence in critical areas of the flight envelope for an eVTOL vehicle such as hover, forward flight, and transitions in between.

Candidate methods will be discussed early on narrowed down to at least two. Approaches to be considered for evaluation and comparison include prior research by REALM for model free / black box / grey box model approaches (e.g. neural certificate framework and ref [1]) as well as those that leverage Gaussian Process regression.

The adaptive envelope protection method should run simultaneously in parallel to any other experimental adaptive/learning control and system identification process for a tiltwing eVTOL aircraft intervening only when needed to ensure safety (i.e. mitigate loss of controllability) or loss of these safety guarantees.

Any approach must assume that a-priori aerodynamic model information is unreliable and thus should minimize reliance on any a-priori model information as much as possible and maximize online utilization of measured/estimated state information (black box model or with model structure assumptions).

The adaptive envelop protection method should be able to dynamically determine and adapt local safe maneuvering envelopes and safe control policies based on the discrepancies between a most current online model and flight data.

Any approach must be demonstrated in a nonlinear software simulation of a tiltwing eVTOL vehicle performing demonstrations representative of flight experiments with a real vehicle. The demonstration should address the hover take-off to forward flight and back to a hover landing transition. If a tiltwing simulation is not available, the study can begin with the most relevant high dimensional nonlinear simulation of an eVTOL vehicle available and transition to simulation of a tiltwing vehicle when it becomes available. The importance of the tiltwing configuration is the changing vehicle configuration as an essential aspect.

Any approach should also:

1) incorporate the practical realities of the real eVTOL vehicle control problem. Examples of such aspects are constrained control inputs, measurement and estimation uncertainty, time delays, actuator uncertainty, disturbances and noise, high dimensional nonlinear model, high relative degree.

2) be adaptable to failure events such as a surface or motor failure and to changes in control allocation (which may be necessary due to a surface or motor failure, or a high temperature motor condition).

3) assume unmodeled dynamics from effects such as rotor-rotor and rotor-aircraft aerodynamics interactions, as well as account for uncertainties in aerodynamic parameters and external disturbances, thrust provided by each rotor, control forces provided by control surfaces, and mass properties of the vehicle.

Implementation and demonstrations in software simulation should always aim to address a realistic flight experiment using an actual tiltwing vehicle to demonstrate safe transitioning, ensured by adaptive envelope protection, from take-off hover to forward flight and back to hover and landing.

Weekly meetings will be held with Dr Chuchu Fan to collaborate and discuss the current research progress.

The desired outcomes of the effort are:

1) implementation of the approaches-methods in source code

2) demonstration and comparison of the approaches-methods in a software simulation of a tiltwing eVTOL vehicle

3) a publishable collaborative report & presentation describing the approaches-methods and presenting the simulation results to be published after the effort (i.e. this effort does not include preparation for publication and submission)

Period of Performance:

Period of performance: 1 year

References:

[1] SABLAS: Learning Safe Control for Black-Box Dynamical Systems, Zengyi Qin, Dawei Sun, Chuchu Fan