Publication

Abstract : Advancements in AI-driven fluid dynamics are opening new frontiers for estimating fluid properties in smart industrial systems, with promising applications in consumer electronic systems ranging from wearable health monitors to smart home appliances. Such methods, initially centered around machine learning (ML) and subsequently deep learning (DL), offer significantly faster and more accurate estimates compared to traditional approximation-based methods. This study presents two novel DL models that predict fluid flow parameters, specifically pressure, and velocity, within flow fields with different geometric profiles. Our first model learns specific fluid behaviors based on distinct geometries, while our second model employs a fusion architecture to generalize across multiple geometries. Based on our extensive evaluation, we observe that individual models estimate pressure and velocity with high accuracy for distinct geometries, while the fusion model generalizes well across multiple geometric profiles, demonstrating greater practical applicability. This research underscores the potential of DL models to improve fluid dynamics prediction, providing a promising alternative that bypasses the need for physics-guided systems. Our approach bridges industrial-level accuracy with consumer needs, enabling more efficient, adaptive, and intelligent products for everyday use.