475589 Petroleum Coke Morphology Mapping: A Mechanistic Approach Using Machine Learning

Sunday, November 13, 2016
Continental 4 & 5 (Hilton San Francisco Union Square)
Pedro Amorim, Russell School of Chemical Engineering, University of Tulsa, Tulsa, OK

Petroleum Coke Morphology Mapping: A Mechanistic Approach Using Machine Learning

Amorim, P.H., University of Tulsa

2nd Year PhD Student

Research Interests:

Machine learning is an established and important technique for introducing intelligent systems to chemical engineering and augment the knowledge of a subject-matter. In the field of big data analytics, a full comprehension of data is easily achievable using this powerful tool, which can analyze a gamma of parameters simultaneously, and tell how they correlate with the outputs that one may be interested in. This feature, which is not found in other modeling techniques, gives an advantage to a machine learning user, since he will get the correct interpretation of variables that have complex association with each other. While machine learning can surely be applied to innumerous research fields (such as process safety, process optimization, or any field that contains big data processing), the application within this particular research, which is being advised by Dr. Keith Wisecarver, is to classify and map the morphology of petroleum coke, according to different chemical and physical parameters. Moreover, the mechanistic approach will give an explanation on why each type of coke morphology is formed. It does so by analyzing suspension stability and rheological properties of the feedstock (which can be considered a suspension) and observing and extracting information out of the liquid crystal carbonaceous mesophase generated by thermolysis and polymerization reactions. Further modeling techniques involving large eddy simulations will be used, as an attempt to simulate the different coke morphologies in flow conditions.

Research Experience:

My research background is a blend of diverse experimental methods and computational work. Initially, I focused in microemulsion systems applied to environmental remediation, and process design in the oil & gas industry, where I worked in collaboration with Petrobras and the Brazilian Association of Oil, Gas, and Renewable Fuels (ANP). Currently, I am focused on the structure and rheology of complex fluids, big data analytics using machine learning, with applications in process design and process monitoring. My current research group, the University of Tulsa Delayed Coking Project (TUDCP), gives me the opportunity to conduct research funded by several national and international oil companies.

Teaching Experience:

As an undergraduate student, I was the teaching assistant of Unit Operation Involving Solids and Fluids at the Federal University of Rio Grande do Norte for one and a half years, during which I had the opportunity to be the lecturer and conduct process design projects. Moreover, I created and was the teacher of an English class for low income students for 6 months, also at the same university.

Teaching Interests:

My teaching interests are in Fluid Mechanics, Process Design (HYSYS, Comsol, Ansys), Computational Fluid Dynamics Applied to Chemical Engineering, Heat Transfer, Numerical Methods for Chemical Engineering, and Process Analysis and Optimization Using Machine Learning.

Future Direction:

As a faculty member, I intend to continue applying machine learning in different research projects which have complex association of variables and also conduct research associated with industry. More specifically, these would be conducted in two main areas: Process analysis & optimization, and Process safety. The former will include two tasks, including analyzing variables extracted from energy applications involving solar thermal energy and biomass and applying algorithms, such as Neural Networks, Support Vector Machines, and Naive Bayes to get a better understanding of the association of operating and chemical parameters; and to optimize the design and performance of the process at the same time. The latter will involve troubleshooting an already existing process using the same techniques previously listed, as a way to improve the process safety and prevent hazardous situations in industrial plants, a topic that can attract a variety of fields. Machine learning has the potential to be one of the main tools for data processing, in a world where computers and chemical processes are getting increasingly interconnected.

Additionally, I will conduct studies of turbulent flow using large eddy simulations simulation models, especially with applications involving simulations of the human body systems (respiratory treat simulations to study allergens deposition, drug delivery simulations), particle agglomeration and deposition in suspensions, and even studies of air and water pollution. These studies will establish research on the fields of medicine, fluid suspensions, and environment. Besides these topics, research involving oil & gas industry will still be in my list of interests. Therefore, because of the different research topics, I see myself establishing an interdisciplinary research network with other faculty members, in a way to increase the scientific collaboration among a variety of fields.


Amorim Valença, P.H., Waturuocha, A., Wisecarver, K. “2D Axisymmetric Temperature Profile Modelling of a Delayed Coking Drum During Pre-Run Warm Up.” Proceedings from Comsol Conference Boston, 2015.

Extended Abstract: File Uploaded