Weatherford’s Adrian Vuyk teamed with Jordan Reynolds (Kalypso) to present the results of a trial of AI/ML in optimizing milling through casing operations by determining the best settings for faster and safer drilling in changing formation conditions. ML was applied to data from past successful and failed casing exit jobs to identify the key factors involved. Open source tools and Python exploratory data analysis across some 18 measured parameters was run in a GPU machine. Particle swarm optimization was found to out-perform gradient-based algorithms that tend to get stuck in local minima. An ‘Edge’ intelligent device talks to Weatherford’s AccuView system onshore. The results are ‘preliminary’ but AI is believed to have enormous potential.
Debbie Rothe showed how Dow Chemical has deployed enterprise
manufacturing intelligence (EMI) dashboards across its
environmental assets. The dashboards show an aerial view of plant with
pop-ups of asset data, highlighting environmental issues. Following
successes at local plants, Dow has now developed a global roll-up
dashboard for its worldwide plants. Management of change was key to the
program and was enabled by engagement with the Prosci community and the Prosci Change Triangle, a
methodology for driving through change in the face of different levels
of opposition.
Robb Bunge (Noble Energy) and Rebecca Thomas (i2k Connect)
began with a timeline for analytics stretching from linear regression
in the 1700s to today’s (or tomorrow’s?) OSDU. Today,
analytics can unveil nonlinear, multi-dimensional relationships between
different data types. Noble leverages I2K Connect’s AI platform
for automated data QC and loading of log curves and other sources,
along with integrated document search from the SPE’s OnePetro and
other repositories. Use cases include pore pressure/geomechanical
studies for drilling safety and logistics efficiency.
Steve Bitar provided an update on ExxonMobil’s ambitious Open process automation (OPA) initiative. This is currently undergoing a two-year (2020-2021) test of OPA standards and components in collaboration with Yokogawa and other partners (notably Aramco and ConocoPhillips). Field trials will come in 2021-2022. Exxon has trialed a prototype at its catalyst testing R&D facility, connect various data sources through a real time bus (Matrikon/OPC-UA) to applications running in a Dell/EMC 'advanced computing platform’. The trial has leveraged the IEC 61499 functional block, a software object used to build applications. Each FB is independent and encapsulates data, variables and programs. FBs can be combined into application. Several vendors involved in the trial (Siemens, Yokogawa, Schneider and Rockwell) were invited to build FBs independently. The resulting test showed that it was possible to assemble a ‘cohesive’ application from four different developer’s function blocks. A written interface description is sufficient to ensure correct use. Intellectual property can be protected via pre-compiled target libraries. Developers can begin to write interoperable function blocks today for licensing on future OPA compliant system. Bitar concluded that ‘the OPA interface will do for FBs what Foundation Fieldbus did for devices!’
Barry Kelley described Koch Industries’ work on leak detection and repair (LDAR). Today this is manual and repetitive requiring complex, physically demanding and potentially hazardous work in industrial settings. Finding quality talent capable of accurately collecting and analyzing the high volumes of data needed for regulatory reporting is a challenge. Koch’s vision is to automate LDAR process using modern sensing technologies and analytics to reduce emissions, improve reporting and optimize operations. An example of the new approach is an automated leak detection sensor that uses Molex Sensorcon wireless emissions detectors and mSyte analytics platform. When the monitor detects a leak, an alert is sent to designated managers who dispatch an LDAR technician to pinpoint the leak using portable instruments such as a VOC sniffer. The system was calibrated with controlled release testing at the EPA’s Research Triangle Park facility in North Carolina. Longer term tests at the Flint Hills (a Koch unit) Sour Lake Olefins facility preceded a full-scale demonstration in process units at the Corpus Christi refinery. Koch is now augmenting the system with multi-level sensors across the plant. Kelley concluded that Koch and the EPA have successfully developed and tested a first-of-its-kind leak detection sensor network that will lead to cleaner air for all. Koch is now seeking regulatory approval for the new approach.
Fakhri Landolsi sees Equinor’s future as ‘robotized and automated’. But today, ‘where do you put your AI/ML investment?’ Beware of the ‘AI return on investment fallacy’. AI models do not deliver value in a ‘linear case by case fashion’ as ‘the time and cost of building a model exceeds most returns from deploying them’. Operators need to build ‘nonlinearities’ in the way ML models are delivered across large, complex organizations. Success will (eventually) come from ‘ubiquitous AI/ML driving machines everywhere’. Success means scaling fast, which can be an organizational challenge and by ‘creating non-linearity’ by solving classes of problems that can be deployed multiple times at very low cost. Enter the Equinor AI/ML ‘chassis’, an AI/ML framework upon which business teams can build and deploy models. ML experts need to automate their work, turning their AI/ML ‘cottage industries’ into factories. Communication is key!
Next year’s IQPC Houston Intelligent
Automation in Oil & Gas is scheduled for 22-23 February 2021.
This article originally appeared in Oil IT Journal 2020 Issue # 2.
For more information or to comment on this topic email here.