On the computational acceleration of aggregating nanoparticle transport models using deep operator networks:...

Area of Science:

• Environmental Science
• Chemical Engineering
• Computational Science

Background:

• Accurate modeling of nanoparticle (NP) transport in porous media is crucial but computationally intensive due to complex processes like aggregation and attachment.
• Existing models struggle with large-scale simulations and fitting applications due to computational demands and differing time scales of involved processes.

Purpose of the Study:

• To develop a computationally efficient model for simulating nanoparticle transport in porous media, incorporating aggregation.
• To accelerate the fitting of experimental data using machine learning approaches.

Main Methods:

• A Deep Operator Networks (DeepONets) model was developed to learn the space-time evolution of NP concentrations, including advection, dispersion, and aggregation.
• The DeepONet was trained on data from conventional NP transport simulations, achieving significant speedups.
• The trained DeepONets model was integrated into a Tandem Neural Network Architecture (TNNA) for direct inference of transport coefficients.

Main Results:

• DeepONets achieved speedups of up to five orders of magnitude compared to conventional simulations.
• Analysis revealed that NP aggregation significantly influences plume dynamics, increasing velocity and spreading.
• The TNNA accelerated the fitting of experimental breakthrough curves by 33% using inferred transport coefficients.

Conclusions:

• DeepONets provide a computationally efficient method for simulating aggregating NP transport in porous media.
• The TNNA enhances the utility of DeepONets by enabling rapid inference of transport parameters.
• These AI-driven tools offer significant advantages for predicting nanoparticle fate and transport in environmental and engineered systems.