Neuro-QFLC: A Hybrid Neural-Fuzzy Controller for Safe, Real-Time EHD Thermal Regulation

Electrohydrodynamic (EHD) cooling provides compact, fanless thermal management; however, its advantages are closely linked to Joule heating and current-density thresholds that may induce dangerous temperature spikes during workload surges. To introduce Neuro-QFLC, a hybrid controller that integrates a neural surrogate for thermo-electrohydrodynamic dynamics with a Quantum-inspired Fuzzy Logic Controller (QFLC) and a constraint-aware safety filter. The neural surrogate predicts the next peak temperature and the amount of uncertainty that can be calibrated. The QFLC encodes rules for interpreting error trends, and a quadratic-program safety layer uses a log barrier and CVaR-based tail cushions to enforce temperature and current limits. In a representative testbed, Neuro-QFLC gets a temperature RMSE of 0.42 K, a peak overshoot of 1.8%, and a violation rate of 0.12%, all while keeping 6.1 K steady-state headroom at a loop latency of about 9.8 ms (with GPU support). This is better than tuned PID, classical FLC, and a neural-only baseline. Calibration makes predictions more reliable (ECE 0.029, Coverage@95% 95.7%), which makes it possible to use principled risk buffers. Stress testing with sensor noise, dropouts, ageing drift, and external heat pulses shows graceful degradation (violations ≤ 0.41%, recovery ≤ 9.5 s). Ablations show that components are needed: taking away the safety QP raises violations by +1.97%, and taking away CVaR raises tail breaches by +0.54%. These results show that Neuro-QFLC is a practical, understandable, and provably safer way to achieve high-performance EHD thermal regulation that can be used in embedded edge deployment.