Skip to main content

AI-enabled ModEx: A scalable AI workflow to efficiently and accurately calibrate PFLOTRAN process models


EMSL Project ID
60592

Abstract

This project will enhance current model-data integration workflows (ModEx) by leveraging recent advances in AI (e.g., scalable deep neural networks). Our improved AI-enabled ModEx workflow will allow us to calibrate PFLOTRAN’s process models faster using multi-physics field datasets. Our research is timely and vital because (1) recent advances in deep learning provide powerful algorithms to calibrate computationally expensive process models efficiently and accurately, (2) trained AI-models can be re-used for inversion rapidly for the given system, with set physics and domains, (3) our team has acquired large-scale multi-modal field datasets needed for model-data integration, and (4) aligns with the BER’s mission needs on ModEx/DigitalTwins workflows as highlighted in the AI4ESP workshop. The project deliverable is an AI code stack that can be driven by EMSL users via Jupyter notebooks and curated datasets for PFLOTRAN model-data integration. A high-impact scientific publication will be developed on our workflow and ModEx dataset. Our digital products will be available to EMSL/BER community through EMSL-github, allowing AI integration into their existing ModEx workflows. The successful outcome of our AI-enabled ModEx will significantly advance model-data integration workflows for model parametrization currently used in BER community. The AI workflow can be modified with minimal changes to extend it to other PFLOTRAN calibration applications (e.g., flow, thermal, reactive-transport). Impact – Our project will provide data analytics tools to EMSL users to incorporate different components of our AI workflow into their own model-data integration pipelines through Tahoma. Our tool will provide a key pipeline capability that is missing at EMSL in the area of integrating data, and process models via machine learning. For this reason, we anticipate that our tool will enable growth in the user program and will bring in new users who will use this tool in the realm ModEx flow and transport such as Prof. Rajaram (JHU), Prof. Misra (TAMU), and Prof. Pinon (UNM).

Project Details

Start Date
2022-10-05
End Date
2023-05-31
Status
Closed

Team

Principal Investigator

Maruti K. Mudunuru
Institution
Pacific Northwest National Laboratory

Team Members

Eusebius Junior Kutsienyo
Institution
Pacific Northwest National Laboratory

Md Lal Mamud
Institution
Pacific Northwest National Laboratory

Glenn Hammond
Institution
Pacific Northwest National Laboratory