Artificial Intelligence Powered Platform for Capillary Microfluidic Design Automation (AI-CMDA)
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Point-of-care (POC) testing has become an essential tool across modern healthcare, environmental monitoring, food safety, and veterinary applications, enabling rapid and cost-effective diagnostics in situations ranging from centralized laboratories to remote and resource-limited locations. Capillary microfluidic chips (CMCs) show tremendous promise in this arena, as they utilize capillary forces to autonomously drive fluid motion and perform a sequence of operations without relying on external power sources or active control. However, designing or redesigning these chips remains a time-consuming and largely manual iterative process, often requiring several weeks to over a year, depending on assay complexity, material compatibility, and fabrication variability. This slow and empirical development restricts the scalability, responsiveness, and broader adoption of CMC-based solutions, especially during urgent health crises or in resource-limited environments, highlighting the critical need for an integrated framework that systematically improves the design process. To bridge this gap, this thesis develops the Artificial Intelligence-Powered Capillary Microfluidic Design Automation Platform (AI-CMDA) that systematically address experimental analysis, predictive modelling, and optimization to form a unified design automation platform for CMCs. Its first module, Capillary Microfluidic Chip Analysis (CMCA), focuses on automating the extraction of fluidic behaviour from experimental videos by applying deep learning-based segmentation techniques to identify the fluid interface position over time. The second module, Capillary Microfluidic Chip fluid path Prediction (CMCP), develops a hybrid predictive framework that couples reduced-order hydraulic–electric circuit models with neural network models trained on experimental and computational data, enabling fast and accurate prediction of fluid motion through complex microfluidic networks, while significantly reducing computational cost compared to full CFD simulations. The third module, Capillary Microfluidic Design Automation (CMDA), integrates these capabilities into an optimization engine that autonomously refines chip geometries to meet user-defined performance targets, using surrogate-assisted gradient-based optimization. Combined, these modules enable fully automated design, prediction, and optimization of CMCs, accelerating the development of next-generation microfluidic devices. The capabilities of the AI-CMDA platform are demonstrated through the design, fabrication, and experimental validation of a Tree-Shaped Concentration Gradient Generator (TCG-CMDA) optimized for minimum inhibitory concentration (MIC) assays. Leveraging AI-CMDA, the entire design–build–test cycle was reduced from several weeks to less than 24 hours. The platform successfully identified optimized chip geometries that achieved targeted dilution profiles with over 95% agreement between predicted and experimentally measured concentration distributions, verified through fluorescent dye assays. Additionally, fluid path predictions achieved an R² of 0.97 relative to experimental time-course data, validating the platform predictive accuracy. Experimental prototypes fabricated from AI-optimized designs consistently outperformed manually designed counterparts in achieving linear and reproducible concentration gradients across multiple independent runs. This case study highlights the ability of AI-CMDA to dramatically compress prototyping timelines, enhance design reproducibility, reduce reliance on trial-and-error and expert intuition, and establish a scalable, automated pathway for capillary microfluidic chip development. By integrating data-driven design, predictive modelling, and optimization into a single streamlined framework, AI-CMDA paves the way for democratizing and accelerating the deployment of next-generation microfluidic systems. These results demonstrate that AI-CMDA can substantially accelerate CMC development while enhancing precision and scalability for high-performance microfluidic applications.