Back close

Measuring Viscosity of Fluids: A Deep Learning Approach Using a CNN-RNN Architecture

Publication Type : Conference Paper

Publisher : ACM Digital Library

Authors : Dr. Vivek Menon, Vishnu Mohan M. S.

Url : https://dl.acm.org/doi/abs/10.1145/3486001.3486232

Campus : Amritapuri

School : Department of Computer Science and Engineering, School of Engineering

Department : Computer Science

Verified : Yes

Year : 2021

Abstract : Estimating fluid properties is an important problem in fluid dynamics that has garnered increasing attention in the machine learning community. The need for non-invasive, real-time measurement of fluid properties such as viscosity, coupled with the underlying complexity of existing mathematical models, has led to a proliferation of a new class of data-driven techniques. Traditional machine learning approaches have focused on estimating viscosity using measurement data of other fluid properties. In this paper, we propose a deep learning framework that can dynamically measure the viscosity of fluids directly from video frames of fluid flow, effectively avoiding the need for any sophisticated equipment for such measurements. We utilize pre-trained Convolutional Neural Networks (CNNs) to extract key features from video frames which are then fed to a Recurrent Neural Network (RNN) to capture the temporal features of the flow data. For implementing and training the composite CNN-RNN model, we created a new dataset of fluid simulations at various viscosity values. We evaluated multiple hybrid models on this dataset and observed that the VGG19+GRU (2 Layers) model showed superior performance in estimating viscosity. Our approach to estimate viscosity can be generalized and extended for the measurement of other fluid properties.

Cite this Research Publication : Vishnu Mohan M. S. & Vivek Menon (2021). Measuring Viscosity of Fluids: A Deep Learning Approach Using a CNN-RNN Architecture. The First International Conference on AI-ML-Systems, Article No.: 12, Pages 1–5, https://doi.org/10.1145/3486001.3486232.

Admissions Apply Now