PRIMS: Physics-Aware AI for Fluid Identification in Microfluidics

A novel AI framework named PRIMS (Physics-guided Representation for Fluid Identification in Multimodal Sensing) has been introduced to enhance fluid identification on devices used in microfluidic settings. This model tackles the issue of ensuring consistent performance despite fluctuations in flow, pressure, and temperature. Unlike traditional learning approaches that consider sensor data as universal features, PRIMS incorporates physical principles into its representation learning and attention strategies. It comprises three key components: Physics-based Token Vectorization, which converts raw signals from Coriolis and pressure sensors into meaningful token embeddings; Physical Component Synthesizer, which addresses viscosity-related interactions among flow, pressure, and density; and Physics-guided Fusion, which utilizes attention-based integration to identify cross-physical correlations. This methodology seeks to boost generalization and interpretability by utilizing fundamental physical laws that influence fluid dynamics. The findings are published in a paper on arXiv (2607.22422).