Data-driven, Learning-based, Adaptive Control of Solid Fuel Ramjet
Abstract
We propose a three-year project with the aim of developing an online, data-driven, learning-based adaptive control system to modulate the thrust generated by solid fuel ramjets in uncertain operating conditions. The project will combine the learning-based adaptive control expertise of the Estimation, Control, and Learning Laboratory (ECLL) at the University of Maryland, Baltimore Country, the high-speed aerothermodynamics multiphysics simulation expertise of the Computational Hypersonics and Nonequilibrium Laboratory (CHANL) at the University of Arizona (UA), and the experimental facilities/experience at the Naval Air Warfare Center Weapons Division (NAWC-WD). The project has the following major objectives:- Develop multi-fidelity computational models of solid fuel ramjets (SFRJ).- Develop, implement, and demonstrate a learning-based adaptive control system to regulate the thrust generated by the solid fuel ramjet and prevent the engine from unstarting in realistic operating conditions.- Develop an integrated multi-physics framework to simulate the flight, control system, and propulsion dynamics simultaneously in realistic operating conditions.- Characterize the actuator bandwidth requirements to arrest and prevent the SFRJ from unstarting. The proposed project will primarily rely on computational modeling and numerical simulations of the solid fuel ramjet and will include extensive software development and numerical experiments to demonstrate the reliable operation of the control system in realistic conditions. The research activities in the project willsimultaneously support the objectives in the companion project proposed by Dr. Alireza Farahmandi and Dr. Brian Reitz in the control research division at NAWC-WD, who will leverage the ongoing SFRJ experiments at the Propulsion Research Lab (PRL) to validate the novel control system experimentally. At the beginning of the project, we will demonstrate adaptive feedback control of a simplified model of an SFRJ with a learning controller to regulate its thrust. A numerical simulation will be used to tune the learning controller s hyperparameters and demonstrate the ability of the control system to adapt to parametric changes in the SFRJ model to demonstrate real-time adaptation. Next, CHANL at UArizona will develop a more sophisticated computational model of the SFRJ that will accurately predict the dynamic behavior as well as simulate unstart in an engine whose configuration will be similar to the SFRJ currentlyunder development and testing at PRL, NAWC-WD. Simultaneously, ECLL at UMBC will develop an integrated framework to simulate the missile flight dynamics with the SFRJ dynamic model developed at CHANL to generate inlet conditions experienced by an engine in real flight conditions. The learning control system will be integrated within the SFRJ computational code to enable arbitrary thrust regulation. We will design several realistic operating scenarios and investigate the performance of the learning control system. The internal structure of the learning controller, including filters, controller parameterization, and learning hyperparameters, will be iteratively refined. At each iteration, we will test the control system s robustness to the operating conditions variations to demonstrate that the learning controller can compensate for uncertain and unknown changes in the system s dynamics experienced in real-world flight conditions.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 15, 2023
- Source ID
- N000142312468
Entities
People
- Ankit Goel
Organizations
- Office of Naval Research
- United States Navy
- University of Maryland, Baltimore County