DAFD is the first microfluidic design automation software that can deliver a user-specified desired performance using machine earning. DAFD is a free, web-based, and open-source software developed by CIDAR lab at Boston University. DAFD capitalizes on machine learning, low-cost rapid prototyping, experimental data-sets, and computer-aided design (CAD) tools to automatically generate the required design of a microfluidic component to deliver the high-level user-specified performance. DAFD eliminates the need for costly and time-consuming design iterations that are currently necessary to create microfluidic devices that deliver an application specific performance. Additionally, DAFD can predict the performance of a given microfluidic design using the automatically trained predictive neural networks generated by DAFD Neural Optimizer.
DAFD currently supports design automation and performance prediction of flow-focusing droplet generation with mineral oil and DI water as the continuous and dispersed phases. Users can specify their desired droplet diameter, generation rate, and design constraints then receive a design delivering the desired performance. Additionally, DAFD Neural Optimizer provides an automated data-to-model machine learning framework that allows users to upload an arbitrary data-set and get optimized predictive neural networks in return, without requiring substantial machine learning knowledge. By making DAFD open-source and developing DAFD Neural Optimizer, we have enabled the microfluidic community to extend the platform with their custom datasets to create further design automation tools for other fluid combinations and/or microfluidic components.
Key Features
-
Machine-learning based performance prediction of microfluidic droplet generation
-
First-of-its-kind design automation tool for droplet microfluidics
-
An open-source automated data-to-model machine learning framework.
Funding
-
NSF Living Computing Project Award #1522074