A Machine Learning and Computer Vision Approach to Rapidly Optimize Multiscale Droplet Generation

Abstract

Generating droplets from a continuous stream of fluid requires precise tuning of a device to find optimized control parameter conditions. It is analytically intractable to compute the necessary control parameter values of a droplet-generating device that produces optimized droplets. Furthermore, as the length scale of the fluid flow changes, the formation physics and optimized conditions that induce flow decomposition into droplets also change. Hence, a single proportional integral derivative controller is too inflexible to optimize devices of different length scales or different control parameters, while classification machine learning techniques take days to train and require millions of droplet images. Therefore, the question is posed, can a single method be created that universally optimizes multiple length-scale droplets using only a few data points and is faster than previous approaches? In this paper, a Bayesian optimization and computer vision feedback loop is designed to quickly and reliably discover the control parameter values that generate optimized droplets within different length-scale devices. This method is demonstrated to converge on optimum parameter values using 60 images in only 2.3 h, 30× faster than previous approaches. Model implementation is demonstrated for two different length-scale devices: a milliscale inkjet device and a microfluidics device.

Machine learning for microfluidic design and control

“…4 ). 95 This process removes the need for domain knowledge and iterative experimental cycles; after initializing the model with 20 pseudo-random datapoints, the ML loop can converge upon a user-defined performance in 60 points total (20 initial and 4 batches of 10 algorithmically-requested points). In Dressler et al , two RL algorithms, Deep-Q Networks (DQNs) and model-free episodic controllers (MFECs) were compared against human performance and each other in the ability to control both laminar flow between two different fluids as well as droplet generation between two immiscible fluids (water in oil emulsions).…”

“…4 Example of an application of machine-guided microfluidic control in its implementation to optimize droplet generation at multiple length scales. 95 (A) After an initial sampling of the parameter space, (B) a small-scale dataset is generated and (C) automatically analyzed using computer vision methods. (D) These results are then fed into a Bayesian decision policy that determines the next set of data to generate.…”

Deep learning detector for high precision monitoring of cell encapsulation statistics in microfluidic droplets

“…31 Likewise, Siemenn et al developed a method employing Bayesian optimization and watershed segmentation with targets of circularity and droplet yield to converge to optimum flow rates. 33 In the second group, studies have focused on cell sorting where the objective is to encapsulate cells and use machine learning approaches to detect individual droplets with cells so that they can be sorted. For example, Howell et al used YOLOv4-tiny to detect cells, beads, and cell doublets in microfluidic droplets and performed ML-assisted sorting.…”

Simple Structural Descriptor Obtained from Symbolic Classification for Predicting the Oxygen Vacancy Defect Formation of Perovskites

“…For example, studies like finding high-conductivity photovoltaic materials, 9 discovering rapid host materials in Li−S batteries, 10 discovering the optimum bromine doping in methylammonium tin-based perovskites, 11 researching the formation and thermal stability of perovskites, 12 and deep mining stable and nontoxic hybrid organic−inorganic perovskites for photovoltaics, 13 can guide researchers to understand the internal mechanism of materials and prepare new high-performance materials. Second, due to the complex preparation process of lithium batteries and solar cells, 14,15 the preparation of these devices by high-throughput machine learning and AI platforms to accelerate devices' development, such as building an ensemble learning platform for the large-scale exploration of new double perovskites, 16 predicting battery end of life from solar off-grid system field data, 17 estimating the remaining charge of Li-ion batteries based on the noise immune state, 18 and rapidly optimizing multiscale droplet generation by computer vision, 19 are also the research focus. In addition, some researchers are working on improving the original machine learning algorithm and proposing new machine learning potential 20 and density functionals 21 to make machine learning more suitable for the research of energy materials and devices.…”