Abstract
Sucker Rod Pumps (SRP) have been extensively utilized in the Duri field Heavy Oil Operations Unit (HOOU) for more than 6000 production wells. Approximately 2000 of these wells are equipped with dynamometer online that generates a daily dynamometer card (DC). Historically, the pump cards evaluation has led to the identification of several mechanical pump issues such as a traveling valve and standing valve leak that directly impact production.
One step of the traditional process to identification of rod pump failure is based on a manual pump card shape analysis performed for individual wells by different engineers throughout production history. To improve efficiency and reliability of shape analysis, Artificial Intelligence-based data analysis has been recently integrated in the oil and gas industry. This article proposes an approach to pump card classification, developed by the Integrated Optimization Decision Support Center, using a modified Case-Based Reasoning or computer reasoning by analogy approach where new problems are solved by comparison to analogous problems solved in the past.
The proposed methodology begins with definition of a reference DC for every known type of mechanical failure. The reference cards define the analogy set. Actual pump cards are then normalized and compared for similarity against each reference card or analogy using Euclidean distance measure between the actual and reference cards. For each actual pump card, the output of this approach is a set of similarity scores which indicate the pump failure type corresponding to references card shape, if any. The analysis is enhanced through the addition of rules based on pump operational parameters that result in specific pump failure signals. The methodology has been verified against DC evaluations from Subject Matter Experts (SME) and is demonstrated to provide robust pump failure signals more efficiently than by manual interpretation of DC for a series of individual wells.
