2025
In this paper, we present a high-fidelity aeroacoustic shape optimization framework that combines large eddy simulation (LES), high-order flux reconstruction (FR), and the gradient-free Mesh Adaptive Direct Search (MADS) algorithm. The objective is to minimize the overall sound pressure level (OASPL) at a near-field observer by directly computing acoustic emissions from unsteady flowfields. To overcome the high computational cost typically associated with such optimization problems, we implement a parallelized version of the MADS algorithm, enabling constant-time optimization iterations regardless of the number of design parameters, assuming sufficient computational resources. We demonstrate the effectiveness of this framework on three complex three-dimensional configurations: an open cavity, tandem cylinders, and a NACA four-digit airfoil. In all cases, significant reductions in noise—up to 13 dB—are achieved while maintaining or improving aerodynamic performance. These results underscore the promise of combining high-order LES with robust, scalable gradient-free optimization for practical aeroacoustic design applications.
2024
In this paper, we demonstrate the feasibility of coupling high order Flux Reconstruction (FR) methods with the gradient free Mesh Adaptive Direct Search (MADS) algorithm for near field aeroacoustic shape optimization at low Reynolds numbers. Using the direct acoustic method to compute overall sound pressure levels (OASPL), we apply this framework to three two dimensional flow problems: flow over an open cavity, flow past tandem cylinders, and flow over a NACA0012 airfoil. Across these cases, the optimization procedure successfully reduces noise levels, up to 16.5 dB in the tandem cylinder case, while preserving or improving aerodynamic performance. For the airfoil, the optimization yields a new shape that maintains the lift coefficient, reduces drag by nearly 25%, and eliminates acoustic emissions at the monitored location. These results highlight the robustness and promise of combining high order accurate numerical solvers with derivative free optimization for low noise aerodynamic design.
2022
In this work, we present a novel time-step-independent filtering approach to stabilize high-order Large Eddy Simulations (LES) using the Flux Reconstruction (FR) method. High-order methods offer increased accuracy for simulating turbulent flows but are prone to non-linear instabilities, particularly in under-resolved scenarios. To address this, we develop and optimize a new class of exponential filters that target high-frequency modes while preserving the accuracy of lower modes. Through extensive numerical testing—over 14,000 simulations—we identify optimal filter parameters that maintain stability across a range of Mach numbers and polynomial degrees. We validate the effectiveness of these filters on benchmark cases, including the Taylor–Green vortex, turbulent channel flow, and a stalled airfoil, demonstrating that they stabilize previously unstable simulations with minimal impact on solution accuracy. This work contributes a computationally efficient and robust stabilization technique suitable for high-order unstructured simulations in complex geometries.
2016
In this work, we focused on improving the energy and cost efficiency of the C3MR liquefaction process, one of the most widely used systems in the liquefied natural gas (LNG) industry. These systems consume a lot of energy, so we applied a combined exergoeconomic analysis—which blends thermodynamic and economic evaluation—to pinpoint inefficiencies in both energy use and cost. We simulated the process using Aspen HYSYS and linked it to MATLAB to apply a genetic algorithm for optimization. Our goal was to maximize the system’s energy efficiency while minimizing the cost of producing LNG. Because these two objectives often conflict, we used multi-objective Pareto optimization to find the best trade-offs. The results showed significant improvements in both efficiency and cost over the original design. We also conducted a sensitivity analysis to see how changes in compressor pressure ratios affect total cost. Overall, this method provides a more effective way to design and operate LNG plants with lower energy use and greater cost-effectiveness.
In this study, we focused on improving how we estimate the viscosity of Iranian crude oils, a key property that influences oil flow and production efficiency. Traditional methods rely heavily on lab experiments or region-specific formulas, which can be costly and inaccurate across different oil fields. To address this, we developed a new multi-hybrid model that combines a type of self-learning neural network called GMDH (Group Method of Data Handling) with a Genetic Algorithm to automatically generate accurate viscosity correlations. We trained our model using extensive laboratory data and tested it across three pressure conditions: below, at, and above the bubble point. Our results showed that the model delivers extremely high accuracy—outperforming seven well-known existing methods—and can be a powerful tool for predicting crude oil viscosity without trial-and-error or heavy lab work. This approach could significantly reduce time and cost in reservoir engineering and production planning.
2015
In this research, we set out to improve the energy efficiency of low-temperature refrigeration systems commonly used in industrial applications like petrochemical plants. Traditional methods for optimizing these systems can be time-consuming and limited in flexibility. So, we developed a hybrid modeling approach that combines a machine learning technique called the Group Method of Data Handling (GMDH) with genetic algorithms and simulation tools like Aspen HYSYS. Our model takes into account key variables such as refrigerant composition and system pressures, and it predicts the optimal settings to minimize power consumption. We tested it on both one- and two-stage cascade cycles, and the results were promising—the model achieved significant energy savings and outperformed conventional optimization methods in accuracy and efficiency. This approach offers a practical and scalable way to design better-performing refrigeration systems with less trial and error.