arXiv subjects

Daniel Stoecklein

Publications and source records attributed to Daniel Stoecklein.

4 recordsLinked to original sources

arXiv2019

Shape-design for stabilizing micro-particles in inertial microfluidic flows

Design of microparticles which stabilize at the centerline of a channel flow when part of a dilute suspension is examined numerically for moderate Reynolds numbers ($10 \le Re \le 80$). Stability metrics for particles with arbitrary shapes are formulated based on linear-stability theory. Particle shape is parametrized by a compact, Non-Uniform Rational B-Spline (NURBS)-based representation. Shape-design is posed as an optimization problem and solved using adaptive Bayesian optimization. We focus on designing particles for maximal stability at the channel-centerline robust to perturbations. Our results indicate that centerline-focusing particles are families of characteristic "fish"/"bottle"/"dumbbell"-like shapes, exhibiting fore-aft asymmetry. A parametric exploration is then performed to identify stable particle-designs at different k's (particle chord-to-channel width ratio) and Re's ($0.1 \le k \le 0.4, 10 \le Re \le 80$). Particles at high-k's and Re's are highly stabilized when compared to those at low-k's and Re's. A comparison of the modified dumbbell designs from the current framework also shows better performance to perturbations in Fluid-Structure Interaction (FSI) when compared to the rod-disk model reported previously (Uspal & Doyle 2014) for low-Re Hele-Shaw flow. We identify basins of attraction around the centerline, which span larger release-angle-ranges and lateral locations (tending to the channel width) for narrower channels, which effectively standardizes the notion of global focusing in such configurations using the current stability-paradigm. The present framework is illustrated for 2D cases and is potentially generalizable to stability in 3D flow-fields. The current formulation is also agnostic to Re and particle/channel geometry which indicates substantial potential for integration with imaging flow-cytometry tools and microfluidic biosensing-assays.

physics.flu-dyn↗

arXiv2018

Flow Shape Design for Microfluidic Devices Using Deep Reinforcement Learning

Microfluidic devices are utilized to control and direct flow behavior in a wide variety of applications, particularly in medical diagnostics. A particularly popular form of microfluidics -- called inertial microfluidic flow sculpting -- involves placing a sequence of pillars to controllably deform an initial flow field into a desired one. Inertial flow sculpting can be formally defined as an inverse problem, where one identifies a sequence of pillars (chosen, with replacement, from a finite set of pillars, each of which produce a specific transformation) whose composite transformation results in a user-defined desired transformation. Endemic to most such problems in engineering, inverse problems are usually quite computationally intractable, with most traditional approaches based on search and optimization strategies. In this paper, we pose this inverse problem as a Reinforcement Learning (RL) problem. We train a DoubleDQN agent to learn from this environment. The results suggest that learning is possible using a DoubleDQN model with the success frequency reaching 90% in 200,000 episodes and the rewards converging. While most of the results are obtained by fixing a particular target flow shape to simplify the learning problem, we later demonstrate how to transfer the learning of an agent based on one target shape to another, i.e. from one design to another and thus be useful for a generic design of a flow shape.

cs.LG↗

arXiv2016

Deep Action Sequence Learning for Causal Shape Transformation

Deep learning became the method of choice in recent year for solving a wide variety of predictive analytics tasks. For sequence prediction, recurrent neural networks (RNN) are often the go-to architecture for exploiting sequential information where the output is dependent on previous computation. However, the dependencies of the computation lie in the latent domain which may not be suitable for certain applications involving the prediction of a step-wise transformation sequence that is dependent on the previous computation only in the visible domain. We propose that a hybrid architecture of convolution neural networks (CNN) and stacked autoencoders (SAE) is sufficient to learn a sequence of actions that nonlinearly transforms an input shape or distribution into a target shape or distribution with the same support. While such a framework can be useful in a variety of problems such as robotic path planning, sequential decision-making in games, and identifying material processing pathways to achieve desired microstructures, the application of the framework is exemplified by the control of fluid deformations in a microfluidic channel by deliberately placing a sequence of pillars. Learning of a multistep topological transform has significant implications for rapid advances in material science and biomedical applications.

cs.LG↗

arXiv2016

Optimization of micropillar sequences for fluid flow sculpting

Inertial fluid flow deformation around pillars in a microchannel is a new method for controlling fluid flow. Sequences of pillars have been shown to produce a rich phase space with a wide variety of flow transformations. Previous work has successfully demonstrated manual design of pillar sequences to achieve desired transformations of the flow cross-section, with experimental validation. However, such a method is not ideal for seeking out complex sculpted shapes as the search space quickly becomes too large for efficient manual discovery. We explore fast, automated optimization methods to solve this problem. We formulate the inertial flow physics in microchannels with different micropillar configurations as a set of state transition matrix operations. These state transition matrices are constructed from experimentally validated streamtraces. This facilitates modeling the effect of a sequence of micropillars as nested matrix-matrix products, which have very efficient numerical implementations. With this new forward model, arbitrary micropillar sequences can be rapidly simulated with various inlet configurations, allowing optimization routines quick access to a large search space. We integrate this framework with the genetic algorithm and showcase its applicability by designing micropillar sequences for various useful transformations. We computationally discover micropillar sequences for complex transformations that are substantially shorter than manually designed sequences. We also determine sequences for novel transformations that were difficult to manually design. Finally, we experimentally validate these computational designs by fabricating devices and comparing predictions with the results from confocal microscopy.

physics.flu-dyn↗