Automated Design for Microfluid Flow Sculpting: Multiresolution Approaches, Efficient Encoding, and CUDA Implementation

Abstract

Sculpting inertial fluid flow using sequences of pillars is a powerful method for flow control in microfluidic devices. Since its recent debut, flow sculpting has been used in novel manufacturing approaches such as microfiber and microparticle design, flow cytometry, and biomedical applications. Most flow sculpting applications can be formulated as an inverse problem of finding a pillar sequence that results in a desired fluid transformation. Manual exploration and design of pillar sequences, while useful, have proven infeasible for finding complex flow transformations. In this work, we extend our automated optimization framework based on genetic algorithms (GAs) to rapidly design micropillar sequences that can generate arbitrary user-defined fluid flow transformations. We design the framework with the following properties: (a) a parameter encoding that respects locality to ensure fast convergence and (b) a multiresolution approach that accelerates convergence while maintaining accuracy. The framework also utilizes graphics processing unit (GPU) architecture via NVIDIA's CUDA for function evaluations. We package this framework in a user-friendly and freely available software suite that enables the larger microfluidics community to utilize these developments. We also demonstrate the framework's capability to rapidly design arbitrary fluid flow shapes across multiple microchannel aspect ratios.

Deep Learning for Flow Sculpting: Insights into Efficient Learning using Scientific Simulation Data

“…Each type of pillar is defined by a specific size and location: four diameters ( D / w  = {0.375, 0.5, 0.625, 0.75} for pillar diameter D and channel width w ), and eight different locations in the channel (, where y / w  = 0.0 is the center of the channel). This library is created by the same experimentally validated simulation technique used by Stoecklein et al 1617. and Lore et al 18,.…”

“…The forward model of simulating a fluid flow shape given a sequence of micropillars follows the experimentally validated method used by Stoecklein et al 1617,. which we briefly outline here.…”

Computational methods and software for the design of inertial microfluidic flow sculpting devices

“…The concept and implementation of inertial fluid flow sculpting via pillar sequences has been previously investigated by the work of Amini et al (Amini et al, 2013(Amini et al, , 2014 and Stoecklein et al (Stoecklein et al, 2014(Stoecklein et al, , 2016(Stoecklein et al, , 2017a. We will briefly elaborate on the generalities of flow…”

“…This drives the need for an automated solution to the inverse problem: designing a micropillar sequence that produces a desired fluid flow shape. To date, there are two automated approaches in literature: heuristic optimization via the Genetic Algorithm (GA) (Stoecklein et al, 2016(Stoecklein et al, , 2017a and deep learning via trained Convolutional Neural Networks (CNN) (Lore et al, 2015). While the GA capably optimized existing microfluidic devices and explored novel flow shapes, there exist a few drawbacks to its use.…”

“…The GA is also a stochastic method, with no guarantee of finding global optima using a finite number of searches. For flow sculpting, this leads to excessive runtime (as much as 2 h), which makes swift design iterations difficult (Stoecklein et al, 2017a). On the other hand, the application of deep learning shown by Lore et al operated with an extremely quick time-to-result (≈ 1 s), but lacked in accuracy comparable to the GA (Lore et al, 2015).…”