CFD-based predictive modeling of multi-perforated dew-point indirect evaporative coolers via multiple linear regression and machine learning
International Communications in Heat and Mass Transfer, cilt.180, sa.P4, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 180 Sayı: P4
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.icheatmasstransfer.2026.112725
- Dergi Adı: International Communications in Heat and Mass Transfer
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO)
- Anahtar Kelimeler: CFD, Decision tree, Dew-point evaporative cooling, Machine learning, Multilayer perceptron, Support vector machine
- Çukurova Üniversitesi Adresli: Evet
Özet
Dew-point indirect evaporative cooling (DPIEC) technology offers a sustainable and energy-efficient alternative to conventional vapor compression systems. However, the high computational cost of accurate numerical models, such as CFD, limits their utility in real-time control and large-scale optimization studies. This study develops surrogate models for a multi-perforated dew-point indirect evaporative cooler using CFD data generated across the investigated ranges of seven jointly varied geometric, operational and psychrometric inputs. The CFD dataset initially comprised 20,480 cases. After excluding cases exhibiting reverse flow, 17,525 cases were retained for model development. Seven multiple linear regression and 32 machine learning models, including decision trees, support vector machines and multilayer perceptrons, were developed to predict five key performance indicators across seven geometric, operational and psychrometric input parameters. The resulting surrogate models are data driven but physically grounded in the experimentally validated CFD dataset. MLR-7 provides an explicit 35-coefficient equation, whereas the nonlinear machine learning models primarily prioritize predictive accuracy. Results indicate that machine learning models significantly outperform linear methods, particularly in capturing highly nonlinear metrics such as the COP and cooling capacity. Among the architectures, the MLP-4 emerged as the most versatile model, achieving near perfect accuracy for dew-point effectiveness, process air outlet temperature and cooling capacity. SVM-6 demonstrated superior performance for water consumption while the DT-7 model proved to be the most reliable for COP estimation. The developed surrogate models reduce performance estimation time from hours to milliseconds, providing a powerful toolkit for real-time HVAC system control, rapid prototyping, and climate specific optimization.