[home] [Personal Program] [Help]

---

Session: Control of heat transfer II
Session starts: Wednesday 05 November, 11:40
Presentation starts: 12:20
Room: Lecture room A

---

Víctor Duro (Universidad Carlos III de Madrid, National Institute for Aerospace Technology)
Juan Alfaro (Universidad Carlos III de Madrid)
Isaac Robledo (Universidad Carlos III de Madrid)
Rodrigo Castellanos (Universidad Carlos III de Madrid)
Carlos Sanmiguel Vila (Universidad Carlos III de Madrid, National Institute for Aerospace Technology)

Abstract:
Introduction: Efficient convective heat transfer is critical in a wide range of industrial and aerospace applications, where optimising turbulent jet mixing can significantly enhance system performance. Advancements in artificial-intelligence (AI) strategies for flow control in free jets have demonstrated the potential of data-driven optimisation to unlock substantial gains in mixing and entrainment [1,2]. These developments pave the way for extending AI control to impinging jets, where the direct interaction between the jet and a surface amplifies both the challenges and the opportunities for heat-transfer enhancement. Pioneering work by B. R. Noack’s group and co-workers [1–4] has combined machine-learning control, genetic programming and, more recently, Bayesian optimisation to discover multi-frequency actuation laws that profoundly reshape the coherent structure of turbulent free jets. However, most of these studies have focused on entrainment proxies, such as centre-line velocity decay or jet width growth, rather than on direct heat-transfer optimisation. Moreover, nearly all prior studies have been limited to free jets. In impinging configurations, the physics become richer and more intricate: stagnation-point heating, wall-jet development, and unsteady recirculation introduce additional levers—and pitfalls—for flow manipulation. Consequently, there is a pressing need for control frameworks that directly optimise wall heat flux and leverage the high-dimensional actuation space offered by multi-jet forcing, while containing actuation costs. In this work,we address these gaps by proposing an open-loop optimisation framework that employs a hybrid genetic algorithm (GA), guided by feedback from a heated thin foil (HTF) acting as a wall heat-flux sensor. The proposed scheme explores a six-dimensional actuation space, defined by the fundamental frequencies of six mini-jets, which induce modulated disturbances in the axial flow via radial injection at the jet exit. Experimental Setup and Control Strategy: Experiments are conducted in a turbulent jet facility, where the exit mass flow rate is fixed at m˙ = 4, g/s using an Alicat ScientificTM M-500SLPM controller. This corresponds to a jet exit velocity of U∞ ≈ 4.7,m/s througha circular nozzle of diameter D = 30,mm, yielding a Reynolds number of Re ≈ 9200. A stainless steel heated thin foil (grade 1.4310, thickness 25μm) is mounted flush with the impingement wall at a stand-off distance of L/D = 2, enabling non-intrusive measurement of wall temperature and local heat flux. Six fast-switching valves are positioned radially at 60◦ intervals around the nozzle exit and connected to d = 2,mm diameter orifices. These inject mini-jets perpendicular to the main jet, introducing a modulated disturbance in the axial flow (see figure 1). Each mini-jet is independently actuated via frequency modulation—defined by actuation frequencies fi, ∀i ∈ [1, 6]—with duty cycles and phase offsets held constant. Mini-jet operation is supplied by a secondary regulated air source, controlled via an Alicat ScientificTM M-50SLPM unit, which continuously monitors pressure, flow rate, and temperature. The pneumatic layout ensures consistent pressure delivery and uniform conditions across all mini-jets, while actively minimising oscillations in the system. Flow uniformity across the range of operating frequencies is validated using hot-wire anemometry, confirming the symmetry and consistency of minijet injection. Wall heat transfer is assessed via infrared thermography, with temperature maps converted into heat flux data using a simplified local unsteady energy balance formulation [5,6]. In parallel, particle image velocimetry (PIV) is employed to capture the flow field dynamics induced by actuation. An in-house hybrid genetic optimiser (HyGO) is employed to maximise the average convective heat transfer of the controlled impinging jet in a circular area of diameter Dopt = 2D, captured by the cost function: Ja = ⟨Nu⟩r − ⟨Nu0⟩r / ⟨Nu0⟩r (1) where ⟨ϕ⟩r represents the integral of magnitude ϕ over a circular domain of radius r = D and Nu0 the Nusselt number distribution of the non-actuated case. The optimised control laws are selected by minimizing a composite cost function that balances the heat transfer enhancement Ja against the energy cost of actuation Jb, similar to [6]. To ensure robust evaluation, each candidate solution (referred to as an individual) is assessed through two independent measurements, and its cost function is assigned only if the relative deviation between the two remains below 5%. This threshold accounts for uncertainties associated with infrared thermography and flow unsteadiness. If the criterion is not met, a third measurement is performed to maximise the number of valid evaluations within each generation. Should the deviation remain above 5% after three attempts, the individual is discarded by assigning it a penalising bad-value, thereby preventing its propagation within the genetic pool. The HyGO algorithm integrates a standard binary-encoded genetic algorithm (GA) with the Downhill Simplex Method (DSM), the latter promoting local refinement and addressing the typical lack of exploitation capabilities in GAs. A total of 10 generations, each comprising 100 individuals, are executed to identify the optimal actuation frequencies. Results: Jet actuation significantly modifies the wall Nusselt number (Nu) distribution, as illustrated in figure 1, which compares the steady-jet condition (all valves open), single-valve actuation, and the non-actuated baseline. After several generations of optimisation and parametric exploration, the algorithm identifies a set of actuation frequencies that outperforms both the no-actuation and steady-jet cases (see figure 2). This control law achieves enhanced and more uniform heat transfer distribution while maintaining moderate energy consumption. The optimised actuation induces a redistribution in the axial flow impacting the wall, broadening the effective heat transfer area. Ultimately, such flow condition is studied with PIV to understand the relation between the developed flow structures and the induced heat transfer structure at the wall, that leads to an optimised convective heat transfer scenario. Conclusions: The preliminary results underscore the potential of genetic algorithms to uncover novel control strategies and demonstrate the feasibility of integrating advanced measurement techniques with optimisation algorithms for enhanced thermal management. In this study, GA successfully identified optimal actuation frequencies that, not only enhanced convective heat transfer, but also maintained energy efficiency. The combined use of infrared thermography and PIV provided a comprehensive assessment of the control effects, reinforcing the capability of data-driven approaches to tackle complex fluid dynamics challenges. Subsequent research will focus on leveraging alternative control parameters, encompassing duty cycle variations, phase modulation, or their combined effects, to refine jet flow behaviour. This approach is expected to produce desirable phase-locked patterns, inducing swirling or other effects that lead to even greater improvements in convective heat transfer performance. This, in turn, will enhance the reliability and extend the operational scope of machine learning-based optimisation frameworks in demanding industrial and aerospace thermal management applications. References [1] Zhou, Y., Fan, D., Zhang, B., Li R. and Noack, B. R. (2020). J. Fluid Mech., 897, A27. [2] Li, Y., Noack, B. R., Wang, T., Maceda, G. Y., Pickering, E., Shaqarin, T. and Tyliszczak, A. (2024). J. Fluid Mech., 991, A5. [3] Jiang, Z., Maceda, G. Y., Li, Y., Shaqarin, T., Gao, N., Noack, B. R. (2024). Phys. Fluids 36 (9): 095126 [4] Wu, Z., Fan, D., Zhou, Y., Li, R., Noack, B. R. (2018). Exp. Fluids 59, 131 [5] Robledo, I. and Alfaro, J. and Sanmiguel Vila, C. and Castellanos, R. (2025). Exp. Therm. Fluid Sci., Accepted. [6] Castellanos, R., Ianiro, A. and Discetti, S. (2023). Appl. Therm. Eng., 230, 120621.