The project aims to develop a learning-based adaptive control system for regulating the thrust of solid-fuel ramjets (SFRJs) under realistic flight conditions and in the presence of model uncertainties. By combining multi-fidelity computational modeling, adaptive control synthesis, and integrated multi-physics simulation, the research seeks to enable real-time thrust regulation and prevent engine unstart events. A simplified quasi-steady SFRJ model, derived from analytical ramjet theory and NASA’s chemical equilibrium analysis, is being validated against experimental data from China Lake and embedded within a Retrospective Cost Adaptive Control (RCAC) framework to demonstrate model-free, learning-based thrust regulation. An integrated simulation environment couples the SFRJ dynamics with flight and control system models to emulate realistic flight scenarios, while benchmark controllers such as LQG and reduced-order models provide performance baselines. Ultimately, the work aims to demonstrate that a controller trained on simplified models can adapt online to high-fidelity, dynamically coupled simulations, thereby paving the way for robust, learning-based propulsion control in next-generation hypersonic systems.
Real-world systems such as drones and robots must operate safely within strict physical limits. Enforcing these state constraints while maintaining stability is a core challenge in control design.
Our lab develops real-time control laws that guarantee both safety (forward invariance of a safe set) and stability (convergence to the desired state). Traditional methods like Control Barrier Functions (CBFs) and Model Predictive Control (MPC) require solving optimization problems at each step—often too slow for practical use.
We introduce a constraint-lifting framework that transforms constrained dynamics into an unconstrained space using smooth, sigmoid-based mappings. This eliminates numerical issues near constraint boundaries and enables explicit, optimization-free controllers.
We also explore convex discrete-time CBF formulations and sigmoid-integral Lyapunov functions for fast, safety-preserving control. These techniques have been demonstrated on nonlinear systems, including bi-copters and attitude-control problems, achieving safe and stable performance in real time.
Flexible structures in aerospace, energy, and mechanical systems, such as wings, blades, panels, and beams, are highly susceptible to vibrations induced by aerodynamic loading, external disturbances, and structural nonlinearities. These vibrations degrade performance, accelerate fatigue, and can trigger instabilities such as flutter. Despite decades of work, practical vibration suppression remains challenging due to modeling uncertainty, actuator/sensor placement constraints, spillover effects, and limited sensing.
Classical vibration control approaches often rely on either high-fidelity physics-based models or fully model-free heuristics, each with clear limitations. High-order models are difficult to identify and control in practice, while purely adaptive schemes must carefully manage robustness and spillover. A key open challenge is developing low-complexity, output-feedback vibration controllers that (i) target dominant vibration modes, (ii) tolerate modeling uncertainty, and (iii) interface naturally with experimentally realizable sensing and actuation.
Our recent work investigates two complementary control paradigms for vibration suppression in cantilevered flexible structures.
Robust low-order optimal control. We developed a frequency-weighted H₂/H∞ output-feedback controller using a low-order linear model identified from high-fidelity finite-element simulations. Frequency-weighting filters are used to explicitly emphasize dominant vibration modes, enabling targeted suppression while maintaining robustness to parameter variations and actuator placement uncertainty. The controller is designed without direct access to the full nonlinear model and is suitable for experimental implementation.
Model-free adaptive output feedback control. We implemented Retrospective Cost Adaptive Control (RCAC) to suppress vibrations caused by unknown disturbances using only measured outputs. Both displacement and acceleration feedback are considered. To mitigate spillover effects commonly associated with acceleration sensing, we design signal-conditioning filters that extract displacement-relevant information while suppressing high-frequency amplification. The resulting controller adapts online without requiring a plant model.