Predicting process and regeneration air conditions in LT3 molecular sieve desiccant wheels using machine learning and regression methods

GÜRÜK A. B., GÜZELEL Y. E., OLMUŞ U., Çerçi K. N., BÜYÜKALACA O.

International Communications in Heat and Mass Transfer, cilt.164, 2025 (SCI-Expanded) identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 164
  • Basım Tarihi: 2025
  • Doi Numarası: 10.1016/j.icheatmasstransfer.2025.108811
  • Dergi Adı: International Communications in Heat and Mass Transfer
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, PASCAL, Aerospace Database, Compendex, INSPEC, Civil Engineering Abstracts
  • Anahtar Kelimeler: Decision tree, Desiccant wheel, LT3 molecular sieve, Machine learning, Multilayer perceptron, Multiple linear regression, Support vector machine
  • Çukurova Üniversitesi Adresli: Evet

Özet

The performance of desiccant wheels is influenced by various design and operational variables. Existing models in the literature for LT3 molecular sieve desiccant wheels often include only a limited number of input parameters or have restricted operating ranges. This study addresses these limitations by generating a dataset encompassing 32,975 scenarios, which was used to develop 192 models using four methods such as multiple linear regression (MLR), decision tree (DT), support vector machine (SVM), and multilayer perceptron (MLP). The models were categorized into two groups to allow for various analyses. Group A models predict the process air outlet conditions of the desiccant wheel, namely temperature (Tpo) and humidity (ωpo). Group B models, in contrast, predict the required regeneration air inlet temperature (Tri) necessary to achieve a desired process air exit humidity (ωpo), while also predicting Tpo. MLP models demonstrated the best accuracy across both groups. In Group A, the model coded as MLP-7 achieved an RMSE of 0.2017 °C for Tpo, and MLP-6 yielded an RMSE of 0.0656 g/kg for ωpo. In Group B, MLP-7 recorded an RMSE of 0.1764 °C for Tpo, while Tri had an RMSE of 0.7892 °C. Additionally, in Group A, the models coded as RS, 3rdCross, DT-28, and SVM-5 also delivered reliable results, while in Group B, the RS, 2ndCross, 3rdCross, DT-28, and SVM-3 models performed well.