SBIR Phase I: Data-Driven Module for Prediction of Materials Physical-Chemical Properties Using Machine Learning

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) project is to develop a novel web application for faster and more accurate estimation of important physical and chemical properties of materials used by engineers and scientists from various academic and industrial disciplines. Oil and gas, mining, nuclear waste management, and environmental consulting companies can use this tool to acquire critical information which are esential to: estimate the oil and gas production rates under different hydraulic fracturing and enhanced oil recovery strategies, minimize the risk of groundwater contamination by acid mine drainage, asses the safety of cement-based engineered barrier systems in nuclear waste repositories, and design the efficient treatment and remediation strategies for contaminated sites. It can also provide necessary parameters for analyzing the long-term effects of geothermal energy usage, evaluating nutrient cycling and pesticide contamination in soil systems in agriculture and food production industry, and quantifying the uncertainties associated with carbon dioxide geological sequestration and storage. This SBIR Phase I project proposes to develop a cloud-based application which will use novel machine learning algorithms and deep learning image processing techniques to turn large volumes of data into valuable parameters used by various industries in their decision-making process. Built on artificial neural network models, minimum viable product version of the new application will target the oil and gas market. It will offer a state-of-the-art data-driven module for early adopters including special core analysis (SCAL) service providers, exploration and production companies, and chemical product suppliers. The customized application will introduce a more computationally-efficient alternative to SCAL and digital rock technology field (i.e. combined physics-based and imaging paradigm) and will accelerate the forecasting process of rock properties. Initially, users will be able to calculate one of the most critical properties of the reservoir rock, i.e. permeability. Customizing machine learning algorithms developed in Phase I for rock permeability estimation to other industries will facilitate entering into other markets. Built on Amazon?s AWS cloud service, it will offer a flexible, computationally-scalable, on-demand and cost-effective solution to the customers by decreasing the upfront and maintenance costs of hardware and software. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.