Technology Focus: Production and Facilities (December 2021)

Abstract

After 9 years of contributing to JPT, it’s time to hang up my spurs. I’ve had the privilege of being exposed to a broad range of topics, technologies, and papers (good, bad, and indifferent). The technological advances I have observed, and the changes in our way of working, have been startling. I wanted to use this opportunity, however, to raise a couple of issues, namely the effect and utility of machine learning (ML) and the usage of point-and-click engineering software. Back in June 2018, I wrote an editorial outlining concerns I held regarding the promise of ML/artificial intelligence to resolve complex physics-based engineering problems. Nothing has changed my opinion since then. There is no doubt that ML has utility in advancing our understanding of certain problems, such as monitoring of large numbers of wells and preempting run-time problems. To claim that ML will provide an accurate forecast of, say, a new multi-phase network flow model, however, would be—let’s be diplomatic here—problematic. In my opinion, ML needs to be applied with surgical precision, not as a blunderbuss shot in the rough direction of a problem with hopes that a pellet will hit a target. It’s still too soon to draw conclusions on the broad utility of ML, but it is unlikely to prove to be the universal cure-all that some have promoted. The best way to summarize the what, where, and how, to apply ML is: It depends. Time will tell, and I remain hopeful but cautious. The other issue is a need to maintain an appreciation of the underlying physics of the problem being modeled. Continuously evolving software has enabled greater overview, visualization, and intricate analysis of ever-more-complex systems. This is a definite boon and provides a level of performance insight hitherto unheard of. Such software, however, still requires a pilot to operate it, and key inputs blunderbuss shot in the rough and physics may be masked, or even hidden, beneath a dazzling interface. Optimization of such systems is also nontrivial, yet this nontrivial task is now made almost pedestrian by a mere click of the mouse. It is so easy to accept software-generated solutions, even though they may turn out to be suboptimal (or even wrong) because of some errant input or a poorly defined parameter. I am a strong proponent of the amazing advances made in software, democratizing and seamlessly unifying other-wise specialist tools and disciplines. But no substitute exists for digging into the problem physics/mechanisms themselves and applying simple pen-and-paper (“back-of-the-envelope”) checks. So, the next time that gas bill arrives, I might suggest keeping the envelope it came in; one never knows when it may prove useful. Recommended additional reading at OnePetro: www.onepetro.org. SPE 203139 - Reduce Cost and Schedule by Developing an Optimum Well Surface Facility Program by Cathy Farina, DyCat Solutions, et al. SPE 206181 - Development of Methods for Top-Down Methane Emission Measurements of Oil and Gas Facilities in an Offshore Environment Using a Miniature Methane Spectrometer and Long-Endurance UAS by Brendan Smith, SeekOps, et al. OTC 30292 - Development of an Assessment Procedure To Predict Local Buckling Behavior Using Probabilistic Method With Finite-Element Techniques for Reeled Pipeline With Additional Features by Sheralia Ufairah Abdullah, McDermott International, et al.