A General Deep Learning Method for Computing Molecular Parameters of a Viscoelastic Constitutive Model by Solving an...

Area of Science:

• Rheology
• Polymer Physics
• Computational Fluid Dynamics

Background:

• Predicting molecular parameters from macroscopic viscoelastic properties is crucial for designing complex fluids.
• Current methods face challenges in accuracy and robustness for diverse applications.

Purpose of the Study:

• To develop a general deep learning method for computing molecular parameters of viscoelastic models by solving inverse problems.
• To validate the accuracy, convergence, and robustness of a deep neural network (DNN)-based solver.

Main Methods:

• A deep neural network (DNN) was employed as a numerical solver for inverse problems.
• The Rolie-Poly model was used to simulate linear and non-linear rheometric properties of entangled polymer solutions.
• The DNN solver's performance was tested against various concentrations and noise levels in stress data.

Main Results:

• The DNN-based solver demonstrated rapid convergence and robustness against minor input data noise.
• Accurate prediction of molecular parameters for monodisperse linear lambda DNA solutions was achieved across wide shear rates and concentrations.
• The method successfully predicted power-law concentration scaling, closely matching experimental estimates.

Conclusions:

• Deep neural networks provide an effective and efficient tool for solving inverse problems in rheology.
• The validated DNN solver accurately computes molecular parameters, advancing molecular and formulation design for complex fluids.
• This approach enhances the predictive capability for material functions based on viscoelastic properties.