Machine Learning Tools in COMSOL Multiphysics

This technical report discusses the integration of artificial intelligence (AI) and machine learning (ML) techniques within COMSOL Multiphysics® for solving inverse problems in continuous inkjet technology. The report outlines a numerical workflow that couples COMSOL with Python ML tools, focusing on predicting ink viscosity from droplet shape. It details the challenges faced in optimizing droplet generation parameters, such as ink properties and nozzle geometry, which directly affect print quality. The report describes the direct problem of simulating inkjet breakup using a 2D-axysymmetric computational fluid dynamics (CFD) model and the subsequent inverse problem of inferring viscosity from droplet shape. It presents the methodology for data generation, pre-processing, and the development of a supervised ML model to establish the relationship between droplet shape and viscosity. The findings illustrate the potential of AI/ML to enhance the efficiency of industrial processes by reducing trial-and-error in optimization tasks.