Improving flow efficiency in micro and mini-channels with offset strip fins: A stacking ensemble technique for Accurate friction factor prediction in steady periodically developed flow

Abstract

A novel machine-learning model for micro and mini-channel friction factor prediction is presented in this article. In recent years, periodic array offset strip fins in micro and mini channels have gained attention due to their compactness and aid in heat transfer in chemical processing and electronic cooling devices. The research uses fin length-to-height ratios below one, and Reynolds numbers between one and six hundred contain 2765 geometrically dimensioned. The friction factor, the most significant barrier to heat transfer channel use, must be accurately estimated to design efficient systems. The Krylov subspace method is implemented using the finite element method using the FEniCS software. Additionally, three machine learning algorithms, Catboost, XGboost, and Random Forest regressors, are implemented to predict the friction factor for periodic steady flow through channels with offset strip fin arrays, and the corresponding prediction performance is analysed. A stacking ensemble model is proposed to enhance the precision of friction factor forecasts. The results demonstrate that the proposed method outperforms empirical correlations in predicting friction factor values for specific performance evaluation metrics with an absolute error of 0.403%, indicating that the model accurately represents the complexity of the data.

Keywords

• Offset strip fin
• Friction factor
• Periodically developed flow
• Machine learning
• Stacking ensemble
• Finite element method