Time to Analysis

I have been watching the first episode of the National Geographic TV series, Genius, about the life and work of Albert Einstein. Among many interesting things I learned from this series was the concept of “Gedankenexperimente” or “thought experiments”. One such thought experiment led Einstein to contemplate how time changes for different clocks, one in the town square and the other on a speeding train. His imagination led him to his general theory of relativity, going beyond Newtonian physics into a new framework of how mass and energy relate and how time and space relate. What an experiment!

But the problem of thought experiments is that you can’t really test them in a physical laboratory, usually because the technology or other environmental factors needed are not yet available. I would like to try such a thought experiment, but I am not going to try to delve into the realm of theoretical physics. My thought experiment is about time and about the availability of data to make better decisions.

Let’s assume for the moment that all relevant data is available for our imaginary data scientist working for our imaginary oil company. I cover this challenge in the article on data marketplaces. Working with his AI coworker, Buddy, the data scientist asks his thought question about how to always maintain field equipment in top working order, to eliminate downtime and nonproductive time, and to respond to field upsets by making the right changes automatically. This goes beyond the responsive maintenance practices of today’s oilfield and even beyond the predictive maintenance practices being tested by leading operators. This goes into the realm of “prescriptive maintenance” where the critical equipment and processes are sustained at their optimum level and respond to changes in their environment. We need Buddy to not only understand best practices, but have a ‘get out of trouble for unpredicted conditions’ skill as well.

Given this thought experiment, the time from data discovery, to data collection and analysis, and finally to action goes to zero, to an automated and correct response. Models have been built from historical data and equipment specs. They have been trained through supervised learning techniques and now are teaching themselves by processing real time data feeds and learning lessons from other human operators and other algorithms working on related assets. Alarms are not for equipment that is about to fail, but for processes that are just beginning to go off course from the anticipated behavior.

Just as in Einstein’s thought experiment, we can’t really put this idea to the test in a real oilfield or remote decision support center, just yet. We are working hard on the “predictive maintenance” stage and making good progress. There are several commercial alternatives in the market today for operators to monitor critical equipment and run predictive maintenance models to help guide field operations. Many operators are interested in going beyond just monitoring critical equipment and for the rest of the process following a “run to failure” strategy. With a lot of hard work, many assets can operate at a 95% or higher uptime. But what if the oilfield could be a four 9s operations (99.99%)? How much value would that bring to the field?

A lot of the technology and tools exist today that will be useful for our “prescriptive maintenance” concept. Field instrumentation is improving on newer equipment. Telecommunications are improving from remote field operations to decision support centers. Concepts like “edge” or “fog” computing allow some (or most) of the calculations to do done in the sensor package or process control unit, so less data has to be sent to the central site. Models (done on the digital twin of the field production process) are getting more accurate and precise. Automation can take the results of the model calculations and translate them into “rules” to control the process. Surveillance from external operations centers (or even monitoring-as-a-service through a commercial agreement with a third-party oilfield service company) provides a “second pair of eyes” on the operations and a platform for simulation of possible alternative rules before an operations decision path is decided. Some human operators (but not all) are getting more comfortable working with this digital environment.

Is this innovation or invention? It is probably a little of both, but we can go back to basics without the constraints of current limitations and do a little creative “design thinking” in our “thought experiment” maybe we can find a way around our current challenges. Maybe the data marketplace is not such a crazy idea. Maybe Buddy, our AI coworker isn’t such a threat. Maybe, just maybe we can cut the time to analysis and decision down to zero, so we can move from responsive to predictive to prescriptive maintenance.

Just think, if we can imagine it, maybe we can build it. If we don’t think it is possible, I am sure it won’t happen.