| Abstract | The present study aims at developing a smart multiphase system for efficiency improvement, loss prevention, and sustainability in industrial processes by leveraging the unprecedented potential of machine learning and artificial intelligence in engineering and scientific applications. To achieve this, we investigate the practicality of using four different machine learning algorithms, namely random forest (RF), support vector machine (SVM), Gaussian process regression (GPR), and artificial neural network (ANN), in constructing a data-driven model for predicting the statistics of discrete particles in a continuous gas flow, with a focus on droplet statistics in twin-fluid sprays. Phase Doppler anemometry is first used for the simultaneous measurement of the size and velocity of spray droplets and the acquired data is then represented by a reduced set of physically-interpretable nondimensional variables following a knowledge-guided data augmentation approach. After that, the ability of the algorithms to learn how to predict the radial distributions of Sauter mean diameter and arithmetic mean velocity of droplets is examined. Finally, to evaluate the reliability of the models constructed, the models are employed to predict the distributions of droplet statistics at experimental conditions outside the learning data sets. It is revealed that the models developed based on GPR and ANN offer superior performances in predicting both the size and velocity of droplets, with low levels of error. The accuracy, complexity, and interpretability of the models are further discussed to assess their capabilities for the control, optimization, and fault diagnostics of smart multiphase systems, wherein the model predictions can be either directly compared with real-time measurements or assimilated into the in-situ simulations of edge-computing devices for a comparison and support provision. |
|---|
