Physics-informed AI/ML of geochemical datasets for characterization, parameterization, and prediction of contaminant transport and remediation processes

Geochemical processes involve complex interactions between fluid and solid constituents critical for understanding contaminant fate, transport, and remediation. The development of reactive geochemical transport (RGT) models predicting these processes from first principles is challenging and typically requires extensive model calibration against observed site data. In addition, existing numerical RGT models overestimate reaction rates in complex heterogeneous flow fields due to inaccurate representation of small-scale reaction dynamics. To resolve these issues, we will apply novel unsupervised and physics-informed (scientific) machine learning methods to develop RGT models that accurately represent geochemical reactions/processes by assimilating existing laboratory and site data. Our proposed work will support optimal remediation control by providing fast, robust, and defensible reduced-order RGT models predicting contaminant fate and transport. We call our framework for characterization and prediction of contaminant transport and remediation processes CHEMML. CHEMML will allow us to embed known physics, geochemistry, observed site data, and existing knowledge (e.g., data from laboratory experiments capturing site conditions or even past experiences from other remediation sites; i.e., transfer learning). The CHEMML models will be capable of representing geochemical processes at site and regional scales and can be embedded in existing Earth System Models (ESMs). CHEMML will be highly computationally efficient and capable of replacing existing RGT numerical simulators in performing optimization of remediation operations. In project Phase I, we will demonstrate the application of CHEMML on contamination sites within the DOE complex (e.g., LANL). Under CHEMML, we will design intuitive graphical-user interfaces and then build streamlined input-output capabilities supporting existing web databases. During the next project phases, CHEMML will be deployed to perform geochemical simulations using cloud computing resources and proprietary data. We will expand our work to focus on Superfund sites and contaminations caused by oil & gas drilling, production, processing, and distribution. We will also facilitate the utilization of CHEMML by providing commercial support, consulting, and services. We will demonstrate CHEMML applicability with representative real-world examples. CHEMML framework will be deployable on a range of computing devices: from handhelds and IoT (Internet of Things) devices to supercomputers and cloud resources. Our vision is that customers will be capable of driving from handhelds and IoT devices computations (analyses, model predictions, etc.) executed on cloud computing resources. After that, the results obtained on the cloud will be propagated back to the customers’ devices to make better informed decisions and evaluate alternative scenarios related to contaminant remediation in real time. Under our commercialization plan, we will also target the application of the CHEMML technology in critical areas important for our nation and economy such as environmental management, climate change, geothermal exploration, and carbon storage.