Implementing asset-wide intelligent digital oilfield (iDOF) solutions is more complex than simply "integrating technology." For iDOF to be truly successful and achieve expectations, concepts from different fields, such as artificial intelligence (AI) or soft computing techniques, must be used to help orchestrate the technology and required functionality to deliver the "intelligence" required to meet the challenges associated with optimizing oil production. Fuzzy logic (FzL) is a type of probabilistic logic that is approximate as opposed to being precise or exact. FzL can have a range of truth values or determine a likelihood of being true.

The process of diagnosis for artificial lift systems has always relied solely on the engineer’s criteria when analyzing data. It is not uncommon for two engineers with similar backgrounds to have, on the basis of the same data, differing opinions of a well problem. It is, however, uncommon that both engineers are either completely incorrect or correct in their judgment, which begs the question as to why this occurs. When traditional binary logic (true or false) is used for well analysis, the outcomes must be either this or that, simply because of the nature of binary logic. Unfortunately, most of the time, the results of well diagnoses, especially when attempting to use a predictive approach, cannot be properly represented with a yes-no approach, which means both engineers would be correct to a certain degree; in FzL, this is called the "degree of truth." FzL incorporates the rationale of having different degrees of truth into the world of well diagnosis and adds tremendous value by raising the issue of different conditions having certain degrees of truth of being in development.

The FzL in this solution was built on a set of worldwide expert rules and converts these rules to their mathematical equivalents. Production operation’s expert rules are used to formulate the conditional statements that comprise FzL (e.g., if the motor temperature exceeds a certain limit, the pump can burn out).

This paper discusses how the AI tool determines trends by calculating a slope for a predetermined period of time for the different sensor signals and compares these sets of trends with pre-established sets of rules. The closeness of the measured trends to the models is expressed as "conditioners" that represent the likelihood of each trend occurring over a period of time.

This technique has proven to be very useful in real-time monitoring for production engineers, helping them to perform fast predictions and pump diagnostics to predict unexpected problems, such as pump wear, gas interference, tubing leaks, viscous forces, and solids-plugged intakes. Several case histories are discussed in this paper. The ultimate goal of this smart flow is to deliver field-level actions that extend pump life and reduce pump downtime.

You can access this article if you purchase or spend a download.