Accelerating the design of lattice structures using machine learning.

Lattices remain an attractive class of structures due to their design versatility; however, rapidly designing lattice structures with tailored or optimal mechanical properties remains a significant challenge. With each added design variable, the design space quickly becomes intractable. To address this challenge, research efforts have sought to combine computational approaches with machine learning (ML)-based approaches to reduce the computational cost of the design process and accelerate mechanical design. While these efforts have made substantial progress, significant challenges remain in (1) building and interpreting the ML-based surrogate models and (2) iteratively and efficiently curating training datasets for optimization tasks. Here, we address the first challenge by combining ML-based surrogate modeling and Shapley additive explanation (SHAP) analysis to interpret the impact of each design variable. We find that our ML-based surrogate models achieve excellent prediction capabilities (R2 > 0.95) and SHAP values aid in uncovering design variables influencing performance. We address the second challenge by utilizing active learning-based methods, such as Bayesian optimization, to explore the design space and report a 5 × reduction in simulations relative to grid-based search. Collectively, these results underscore the value of building intelligent design systems that leverage ML-based methods for uncovering key design variables and accelerating design.

The multivariate interaction between Au and TiO2 colloids: the role of surface potential, concentration, and defects.

The established DLVO theory explains colloidal stability by the electrostatic repulsion between electrical double layers. While the routinely measured zeta potential can estimate the charges of double layers, it is only an average surface property which might deviate from the local environment. Moreover, other factors such as the ionic strength and the presence of defects should also be considered. To investigate this multivariate problem, here we model the interaction between a negatively charged Au particle and a negatively charged TiO2 surface containing positive/neutral defects (e.g. surface hydroxyls) based on the finite element method, over 6000 conditions of these 6 parameters: VPart (particle potential), VSurf (surface potential), VDef (defect potential), DD (defect density), Conc (salt concentration), and R (particle radius). Using logistic regression, the relative importance of these factors is determined: VSurf > VPart > DD > Conc > R > VDef, which agrees with the conventional wisdom that the surface (and zeta) potential is indeed the most decisive descriptor for colloidal interactions, and the salt concentration is also important for charge screening. However, when defects are present, it appears that their density is more influential than their potential. To predict the fate of interactions more confidently with all the factors, we train a support vector machine (SVM) with the simulation data, which achieves 97% accuracy in determining whether adsorption is favorable on the support. The trained SVM including a graphical user interface for querying the prediction is freely available online for comparing with other materials and models. We anticipate that our model can stimulate further colloidal studies examining the importance of the local environment, while simultaneously considering multiple factors.

Comparing Direct and Pulsed-Direct Current Electrophoretic Deposition on Neural Electrodes: Deposition Mechanism and Functional Influence.

Electrophoretic deposition (EPD) of platinum nanoparticles (PtNPs) on platinum-iridium (Pt-Ir) neural electrode surfaces is a promising strategy to tune the impedance of electrodes implanted for deep brain stimulation in various neurological disorders such as advanced Parkinson's disease and dystonia. However, previous results are contradicting as impedance reduction was observed on flat samples while in three-dimensional (3D) structures, an increase in impedance was observed. Hence, defined correlations between coating properties and impedance are to date not fully understood. In this work, the influence of direct current (DC) and pulsed-DC electric fields on NP deposition is systematically compared and clear correlations between surface coating homogeneity and in vitro impedance are established. The ligand-free NPs were synthesized via pulsed laser processing in liquid, yielding monomodal particle size distributions, verified by analytical disk centrifugation (ADC). Deposits formed were quantified by UV-vis supernatant analysis and further characterized by scanning electron microscopy (SEM) with semiautomated interparticle distance analyses. Our findings reveal that pulsed-DC electric fields yield more ordered surface coatings with a lower abundance of particle assemblates, while DC fields produce coatings with more pronounced aggregation. Impedance measurements further highlight that impedance of the corresponding electrodes is significantly reduced in the case of more ordered coatings realized by pulsed-DC depositions. We attribute this phenomenon to the higher active surface area of the adsorbed NPs in homogeneous coatings and the reduced particle-electrode electrical contact in NP assemblates. These results provide insight for the efficient EPD of bare metal NPs on micron-sized surfaces for biomedical applications in neuroscience and correlate coating homogeneity with in vitro functionality.

Two Photon lithography additive manufacturing: Video dataset of parameter sweep of light dosages, photo-curable resins, and structures.

This document describes the collection and organization of a dataset that consists of raw videos and extracted sub-images from video frames of a promising additive manufacturing technique called Two-Photon Lithography (TPL).  Four unprocessed videos were collected, with each video capturing the printing process of different combinations of 3D parts on different photoresists at varying light dosages.  These videos were further trimmed to obtain short clips that are organized by experimental parameters. Additionally, this dataset also contains a python script to reproduce an organized directory of cropped video frames extracted from the trimmed videos. These cropped frames focus on a region of interest around the parts being printed. We envision that the raw videos and cropped frames provided in this dataset will be used to train various computer vision and machine learning algorithms for applications such as object segmentation and localization applications. The cropped video frames were manually labelled by an expert to determine the quality of the printed part and for printing parameter optimization as presented in "Automated Detection of Part Quality during Two-Photon Lithography via Deep Learning" [1].

Automated detection and sorting of microencapsulation via machine learning.

Microfluidic-based microencapsulation requires significant oversight to prevent material and quality loss due to sporadic disruptions in fluid flow that routinely arise. State-of-the-art microcapsule production is laborious and relies on experts to monitor the process, e.g. through a microscope. Unnoticed defects diminish the quality of collected material and/or may cause irreversible clogging. To address these issues, we developed an automated monitoring and sorting system that operates on consumer-grade hardware in real-time. Using human-labeled microscope images acquired during typical operation, we train a convolutional neural network that assesses microencapsulation. Based on output from the machine learning algorithm, an integrated valving system collects desirable microcapsules or diverts waste material accordingly. Although the system notifies operators to make necessary adjustments to restore microencapsulation, we can extend the system to automate corrections. Since microfluidic-based production platforms customarily collect image and sensor data, machine learning can help to scale up and improve microfluidic techniques beyond microencapsulation.

Experimental characterization and modeling of optical tweezer particle handling dynamics.

We report a new framework for a quantitative understanding of optical trapping (OT) particle handling dynamics. We present a novel three-dimensional particle-based model that includes optical, hydrodynamic, and inter-particle forces. This semi-empirical colloid model is based on an open-source simulation code known as LAMMPS (large-scale atomic/molecular massively parallel simulator) and properly recapitulates the full OT force profile beyond the typical linear approximations valid near the trap center. Simulations are carried out with typical system parameters relevant for our experimental holographic optical trapping (HOT) system, including varied particle sizes, trap movement speeds, and beam powers. Furthermore, we present a new experimental method for measuring both the stable and metastable boundaries of the optical force profile to inform or validate the model's underlying force profile. We show that our framework is a powerful tool for accurately predicting particle behavior in a practical experimental OT setup and can be used to characterize and predict particle handling dynamics within any arbitrary OT force profile.

Mesoscale Particle-Based Model of Electrophoretic Deposition.

We present and evaluate a semiempirical particle-based model of electrophoretic deposition using extensive mesoscale simulations. We analyze particle configurations in order to observe how colloids accumulate at the electrode and arrange into deposits. In agreement with existing continuum models, the thickness of the deposit increases linearly in time during deposition. Resulting colloidal deposits exhibit a transition between highly ordered and bulk disordered regions that can give rise to an appreciable density gradient under certain simulated conditions. The overall volume fraction increases and falls within a narrow range as the driving force due to the electric field increases and repulsive intercolloidal interactions decrease. We postulate ordering and stacking within the initial layer(s) dramatically impacts the microstructure of the deposits. We find a combination of parameters, i.e., electric field and suspension properties, whose interplay enhances colloidal ordering beyond the commonly known approach of only reducing the driving force.

Electric double-layer structure in primitive model electrolytes: comparing molecular dynamics with local-density approximations.

We evaluate the accuracy of local-density approximations (LDAs) using explicit molecular dynamics simulations of binary electrolytes comprised of equisized ions in an implicit solvent. The Bikerman LDA, which considers ions to occupy a lattice, poorly captures excluded volume interactions between primitive model ions. Instead, LDAs based on the Carnahan-Starling (CS) hard-sphere equation of state capture simulated values of ideal and excess chemical potential profiles extremely well, as well as the relationship between surface charge density and electrostatic potential. Excellent agreement between the EDL capacitances predicted by CS-LDAs and computed in molecular simulations is found even in systems where ion correlations drive strong density and free charge oscillations within the EDL, despite the inability of LDAs to capture the oscillations in the detailed EDL profiles.

Model-free test of local-density mean-field behavior in electric double layers.

We derive a self-similarity criterion that must hold if a planar electric double layer (EDL) can be captured by a local-density approximation (LDA), without specifying any specific LDA. Our procedure generates a similarity coordinate from EDL profiles (measured or computed), and all LDA EDL profiles for a given electrolyte must collapse onto a master curve when plotted against this similarity coordinate. Noncollapsing profiles imply the inability of any LDA theory to capture EDLs in that electrolyte. We demonstrate our approach with molecular simulations, which reveal dilute electrolytes to collapse onto a single curve, and semidilute ions to collapse onto curves specific to each electrolyte, except where size-induced correlations arise.