Ensemble latent assimilation with deep learning surrogate model: application to drop interaction in microfluidics device

A major challenge in the field of microfluidics is to predict and monitor drop interactions. In this work, based on experiments performed on a microfluidics device, we developed an image-based data-driven model to forecast drop dynamics. To alleviate the computational burden, reduced-order modelling techniques, namely proper orthogonal decomposition and convolutional neural networks, are applied to compress the recorded images into low-dimensional spaces. Recurrent neural networks are then employed to build a surrogate model of drop interactions by learning the dynamics of compressed variables in the reduced-order space. To incorporate real-time observations, we developed an ensemble-based Latent Assimilation algorithm scheme. With the help of ensemble-based data assimilation techniques, the novel approach improve the prediction results by adjusting the starting point of the next time-level forecast. The performance of the developed system is evaluated by experimental data (i.e., recorded videos), which is excluded from the training of reduced-order modelling and recurrent neural networks. The developed scheme is general, and it can be applied to other dynamical systems.