Within the framework of the 18th edition of the Conference on Data Science & Artificial Intelligence organized by the Faculty of Engineering of the Universidad AustralA number of companies shared how they are applying cutting-edge technology to transform their organizations. 

In this regard, some use cases in the OIL&GAS industry carried out in YPF and Pan American Energy were highlighted, which propose improvements in the processes.  

Machine Learning use cases in the OIL&GAS industry

Gonzalo Mognol, Data Scientist at YPF, shared a project that aimed detect anomalies in torches, These are safety features in the refinery where process waste is burned. In order to move beyond passive monitoring of the flares (They are monitored 24 hours a day for security reasons) Video analytics was used with infrared spectrum cameras.  

How was the project carried out? An algorithm was developed to detect a large flame in any weather condition: First, the large flame was detected as an object. Then, the bounding box area (the rectangle surrounding the detected object) was identified. Finally, the flame's outline was defined using OpenCV algorithms. This allowed researchers to obtain information about the flame's position and the conditions necessary for the high flame to occur. Using this technology, they trigger email alerts when a high noise event and high emissions of black smoke occur, generally associated with high flame events. 

For its part, Johann Cambra, also a Data Scientist at YPF, shared another project linked to understanding the soil in Vaca Muerta. With the aim of improving the petrophysical model of Vaca Muerta, Argentina's largest oil and gas reservoir, YPF developed models to understand the subsoil and locate where the oil is.  

A petrophysical model seeks to understand the subsurface, which is composed of pores (where gas, oil, and water are found) and a matrix (which gives rise to the pores). To generate reliable models that simplify the workflow, Data from 26 wells and laboratory data were used. This allowed the company to obtain information on the areas with the greatest potential for oil extraction and, in addition, to save future expenses when acquiring new information. 

Analytics applied to the Oil & Gas industry. The case of Pan American Energy

Engineer Alejandro Andrés Bulgheroni, leader of the digital transformation process at PAE, He shared the model the company uses to carry out data science-based projects with the goal of optimize the company's operational processes.  

Bulgheroni shared a series of algorithms that they have developed at Pan American Energy, such as:  

• The dynamometer chart classifier that allows generating an operational diagnosis of mechanical pumping wells using artificial intelligence algorithms for image classification.  
• A time-to-failure algorithm for mechanical pumping (a Random Forest Survival algorithm was implemented) allows assigning a failure risk to wells with mechanical pumping based on their characteristics, operating conditions, and accumulated fatigue of the equipment.  
• A model based on physical equations to determine the submergence of wells with PCP pumps (Progressive Cavity Pumps) based on the torque reading and other variables in the surface sensors.  
• The Digital Well Assistant: a well monitoring system that, based on (i) the static characteristics of the well, (ii) dynamic readings from the well's sensors, and (iii) the results of physical, artificial intelligence, and machine learning models such as the dynamometer chart classification, the time-to-failure algorithm, or the PCP bottomhole pressure model mentioned above, generates an operational recommendation aimed at optimizing production and reducing the risk of failure.  
• An operational research algorithm for optimizing the Pumping Equipment (PE) fleet  

The main challenge they face is getting the algorithms adopted effectively, impacting operational processes. To achieve this, it is critical to bring together 3 areas of knowledge: