Problem Statement

Practical wind-tunnel-in-the-loop DRL requires reliable reward computation from limited pressure taps rather than full-field measurements.

Why It Matters

Quantifying where pressure-only sensing succeeds or fails is important for experimental AFC reward design and sensor budgeting.

Methods / Numerical + ML Setup

• Studied a 3D NACA0024 airfoil at 8° incidence and chord-based Reynolds number 60,000.
• Generated time-resolved pressure fields for uncontrolled and actuated cases using wall-resolved implicit LES in PyFR.
• Trained regression models to reconstruct lift and quarter-chord pitching moment from sparse pressure signals over DRL-relevant control intervals.
• Formulated sensor placement as a constrained optimization problem with a fixed pressure-tap budget.

Key Results

• Established reconstruction-error bounds across candidate sensor budgets.
• Derived practical placement guidelines for pressure taps in future DRL wind-tunnel campaigns.

Toolchain

• PyFR
• Python
• Regression modeling
• Optimization workflows
• Active flow control setup

Challenges and Lessons

• Balancing reconstruction accuracy against strict sensor-count constraints.
• Maintaining robust estimates under unsteady actuation and transitional flow behavior.

Future Work

• Integrate optimized tap layouts into real-time reward computation during closed-loop DRL experiments.
• Extend sensor-placement studies across broader operating envelopes and control objectives.