Machine_Learning_Technique_For_Pressure_Drop_Prediction

Two Phase Pressure drop in a straight horizontal pipe

Image Credit: Mudawar et. al.

Motivation

• Two-phase pressure drop prediction of refrigerant fluids is very essential in many cooling application. In the past many decades thermal system design engineers solely relied on many empirical and semi-empirical pressure drop models which are developed for specific fluids and narrow range of operating conditions, this makes prediction accuracy poorer.

• In the present study the prediction of two-phase frictional pressure drop of refrigerant fluid R22, R410a, R134a in a horizontal straight pipe during adiabatic wall conditions is made with the application of Artificial Neural Networks (ANN) Random Forest Regressor (RFR), and Xtreme Gradient Boosting Regressor (XGBR).

Experimental Flow Visualization (Padilla 2011 et al )

Two Phase Heat Transfer in a straight horizontal pipe

Image Credit: Mudawar et. al.

Methods

• More than 1000+ experimental two-phase pressure drop data is taken from the literature with features- mass flux, vapor quality, pipe diameter, saturation pressure and saturation temperature.

• Three different ANN, RFR and XGBR models trained with 22 input features to predict output pressure gradient.

Conclusion

• Among ANN, RFR, and XGBR, It is concluded that the application of ANN is very powerful tool when it comes to accurate of prediction two-phase thermo-hydraulics of refrigerant pipes.