Dr. John Hedengren of Brigham Young University, Provo, Utah

A convergence of several key technologies creates an opportunity to use sophisticated mathematical models within automation. A significant challenge is the size of the physics-based models that have too many adjustable parameters or are too slow in simulation to extract actionable information. This presentation shows how fit-for-purpose models can be used directly in automation and optimization solutions. These fit-for-purpose models have unlocked new ways of thinking. Hybrid modeling uses the strengths of both physics-based and data-informed modeling approaches. A hybrid approach uses a priori knowledge in the form of a nonlinear physics-based model with empirical model elements. The challenges and opportunities for combining physics-based and data-driven elements are discussed. The methods are demonstrated on two applications: mineral processing and drilling automation.

An important mineral processing application is a fluidized bed roaster for gold ore processing. Refractory carbonaceous ore contains naturally occurring material that negatively affects the cyanidation gold recovery process by encapsulating the solubilized gold. The roaster oxidizes these carbonaceous materials. The reaction kinetics of the fluidized bed reactor are modeled with a shrinking core model (SCM) in conjunction with the heat and material balance equations. The unknown parameters in the model are effective diffusivity, particle size, and heat transfer coefficients that are empirically estimated with operation data. The hybrid physics and machine learning model is used in a real-time optimization The hybrid model demonstrates higher accuracy over traditional physics-based or empirical neural net approaches.

The drilling industry faces challenging market conditions that motivate the use of automation to reduce costs and decrease well manufacturing variability. Physics-based and empirical models have advantages and opportunities for predictive monitoring and control. Current progress, challenges, and opportunities to control critical drilling conditions such as downhole pressure in Managed Pressure Drilling (MPD) are explored. In automated rig systems, there is additional potential to unlock the predictive capabilities of physics-based models to "see" into the near future to optimize and coordinate control actions.

Biography:

Dr. Hedengren received a PhD degree in Chemical Engineering from the University of Texas at Austin and is an Associate Professor at Brigham Young University. Previously, he previously worked with ExxonMobil on Advanced Process Control. His primary research focuses on accelerating machine learning and automation technology across industries. Other research interests include fiber optic monitoring, Intelli-fields, reservoir optimization, drilling automation, nuclear hybrid energy systems, and unmanned aerial systems. He is a leader of the Center for Unmanned Aircraft Systems (C-UAS), applying UAV automation and optimization technology to energy infrastructure.