Turkish Journal of Electrical Engineering and Computer Sciences
Author ORCID Identifier
GÖKHAN DEVECİ: 0009-0007-7288-836X
ÖZGÜN YÜCEL: 0000-0001-8916-2628
ALİ OLCAY: 0000-0003-0995-9173
Abstract
While Reynolds-averaged Navier–Stokes (RANS) simulations are essential for combustion modeling, their high computational cost creates a significant design bottleneck. This bottleneck is amplified when using tunable turbulence models like GEKO; determining the optimal value for parameters such as Cmix requires an iterative process of running numerous, time-consuming simulations, making traditional parametric analysis impractical. This study presents a novel, machine learning (ML)-based software methodology, the ‘Mean Temperature Prediction Tool’, developed to eliminate this bottleneck by providing instantaneous predictions of mean temperatures in swirl-stabilized hydrogen-methane flames. A computational fluid dynamics (CFD) database comprising 225 unique design points was generated using Latin hypercube sampling. This database was used to train support vector regression (SVR), random forest, and extreme gradient boosting (XGBoost) ML models to map five input parameters to mean temperature outputs at seven distinct axial x/D locations. The developed software encapsulates these ML models to generate instantaneous ‘Temperature Distribution’ heatmaps and ‘Sensitivity Maps’ (∂T/∂Cmix ), allowing for transparent and rapid analysis of parameter influence. Model performance evaluations confirmed high overall accuracy (R2 > 0.90); XGBoost demonstrated the most consistent high performance, while SVR achieved the highest single-point accuracy (R2 = 0.99). As a final validation, the tool was benchmarked against the SMH-1 and SMH-2 cases. Rather than relying on qualitative visual agreement, quantitative metrics confirmed the tool’s exceptional surrogate fidelity, achieving a mean absolute percentage error of just 0.58% and a Normalized RMSE of 0.69% against the reference CFD solutions.
DOI
10.55730/1300-0632.4293
Keywords
Generalized k − ω (GEKO) turbulence model, Cmix, methane-hydrogen combustion, sensitivity analysis
First Page
1126
Last Page
1149
Publisher
The Scientific and Technological Research Council of Türkiye (TÜBİTAK)
Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
DEVECİ, G, YÜCEL, Ö, & OLCAY, A. B (2026). A machine learning-based software tool for Cmix tuning and temperature sensitivity analysis in generalized k − ω hydrogen-methane flames. Turkish Journal of Electrical Engineering and Computer Sciences 34 (6): 1126-1149. https://doi.org/10.55730/1300-0632.4293
Included in
Computer Engineering Commons, Computer Sciences Commons, Electrical and Computer Engineering Commons