Predicting buoyant jet characteristics: a machine learning approach.

Area of Science:

• Fluid dynamics
• Machine learning applications

Background:

• Positively buoyant miscible jets are crucial in various industrial and environmental processes.
• Accurate prediction of jet characteristics, such as laminar length and spread angle, is essential for modeling and control.

Purpose of the Study:

• To employ supervised machine learning techniques for predicting key characteristics of miscible jets.
• To compare the performance of different machine learning algorithms against empirical correlations.

Main Methods:

• High-speed imaging and planar laser-induced fluorescence were used for data acquisition.
• Supervised machine learning algorithms including linear regression, support vector regression, random forests, K-nearest neighbour, and artificial neural networks were utilized.
• Model performance was evaluated using standard metrics.

Main Results:

• The random forest algorithm demonstrated superior performance in predicting jet characteristics compared to other tested models.
• The random forest model significantly outperformed a recent empirical correlation, particularly in predicting laminar length.
• Accurate predictions were achieved across a wide range of Reynolds and Archimedes numbers.

Conclusions:

• Supervised machine learning, specifically the random forest algorithm, offers a highly accurate method for predicting miscible jet characteristics.
• This approach provides a significant improvement over traditional empirical correlations, especially for laminar length prediction.
• The findings have implications for enhanced modeling and control of buoyant jet phenomena.