Paper Number
AM5
Session
AI and ML Based Rheological Characterization
Title
Data-driven selection of thixotropic models via Rheology-informed Neural Networks (RhINNs)
Presentation Date and Time
October 10, 2022 (Monday) 11:10
Track / Room
Track 6 / Mayfair
Authors
- Saadat, Milad (Northeastern University)
- Mahmoudabadbozchelou, Mohammadamin (Northeastern University)
- Jamali, Safa (Northeastern University)
Author and Affiliation Lines
Milad Saadat, Mohammadamin Mahmoudabadbozchelou and Safa Jamali
Northeastern University, Boston, MA
Speaker / Presenter
Saadat, Milad
Keywords
computational methods; AI based; ML based
Text of Abstract
A myriad of empirical and phenomenological constitutive models have been proposed and developed to represent a variety of responses of complex fluids. As the material response gets more sophisticated, more complex models are inevitably called for, with their own advantages and limitations. In selecting appropriate models for a given data set, automatic algorithms that eliminate user biases are of particular interest, especially when such models have an increasing number of fitting parameters, making the parameter recovery a non-trivial task. Here, we present a rheology-informed neural network (RhINN) that enables robust model selection based on available experimental data with minimal user intervention. Following our previous work, in which an automated platform was introduced to select steady-state models from a given library, here, we focus on transient flows and embed several thixotropic models into our algorithm. We train our RhINN using a series of transient experimental data and ask the algorithm to select the most accurate thixotropic model [with the fewest fitting parameters] capable of describing the observed rheology. We will show that RhINN can expeditiously select the best model and recover the model parameters for various transient data sets, a task that can be challenging for thixotropic models with a prohibitive number of fitting parameters.