Abstract
Sucker-rod pumps (SRP) are the most common form of artificial lift in depleted oil wells. Given vast datasets collected from years of operation, many operators are enacting digital technology to generate automated artificial lift systems. However, their online monitoring and optimization come with many challenges. Therefore, the individually engineered artificial lift is an imminent solution for long-term production, while maintaining cost efficiency. The key to make sucker-rod pumps operate effectively lies in downhole condition diagnostics. The emerging big-data analytics have provided relatively precise downhole condition forecasting based on available data, enabling better decision making. This study focuses on developing a testing digital SRP application, while leveraging analytical approaches to diagnose its operational anomalies.
This study presents an experimental and analytical workflow to monitor sucker-rod pumps and perform diagnostics using a designed Interactive Digital Sucker-Rod Pumping Unit (IDSRP). This unit consists of a vertical 50-ft long facility with a downhole rod pump at the bottom and proper instrumentation, capable of simulating a rod-pumped wellbore's operation. A linear actuator is used to provide the rod string's reciprocating movement and simulate different surface units and operating scenarios. The system utilizes the application of Pulse-width Modulation (PWM) technique and data-acquisition system (DAQ) to obtain analog results through signals detected by sensor. The surface dynamometer cards, and time-driven pressure and rate data are collected to train a cloud-based analytics software platform. The wave equation of Gibbs is used to draw the downhole pump card from the surface card. Some of the tested scenarios are normal pump operation with varying rates and varying levels of pumping off at the pump inlet.
The applied online prototype is designed to provide a step towards digitized automation systems. The setup is used to generate datasets for the rod-pump's operation at varying pump speeds, stroke lengths, and rod movement profiles. The collected data include the flowrate, bottomhole and surface pressures, and the dyno cards. The digitized transformation algorithms develop these physics-based inputs to generate predictive models, thus classifying operational conditions or failures of the pump. The model dynamically categorizes the pumps into key states of ideal condition and over-pumping with a regression fit accuracy of higher than 0.7 and overall classification accuracy of 92%.
The novelty of this setup consists not only of its mechatronic design but also allows thorough performance monitoring of the pump, thus easily validating models. The results have the potential of becoming a tool to optimize and shorten the downtime for repairing pump failures.