PCP Event Detection via ML Anomaly Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Oil and gas wells often operate unattended in remote areas, leading to challenges in detecting abnormal pump operations and reducing downtime, which results in lost productivity, increased costs, and safety risks for field personnel.
Innovation Solution
The implementation of a machine learning (ML) based event detector that monitors and controls progressing cavity pump (PCP) operations by using anomaly detection to identify abnormal operations, providing explanations for the anomalies, and taking predefined corrective actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If field personnel physically inspect equipment at remote well sites, then they can detect and repair problems, but this results in loss of productivity and profitability while also creating safety risks
Solution Approach 1:
The system enables self-monitoring and self-diagnosis of PCP equipment through automated sensors and machine learning algorithms that detect anomalies and predict failures without human intervention, allowing the equipment to essentially monitor and report on its own operational status
Solution Approach 2:
Physical inspection by field personnel is replaced with an automated electronic monitoring system using sensors, data transmission, and machine learning algorithms to detect equipment anomalies and predict failures remotely
2Reliability
If field personnel travel to well sites for physical inspection, then problems can be addressed, but this is costly and time-consuming
Solution Approach 1:
The system performs preliminary detection and diagnosis of equipment problems through continuous monitoring and anomaly detection algorithms, identifying potential failures before they occur and enabling proactive maintenance scheduling
Solution Approach 2:
The system continuously collects operational data from sensors, analyzes it through machine learning models, and provides real-time feedback on equipment status, enabling rapid response to detected anomalies without requiring physical site visits
3Measurement precision
If traditional monitoring methods are used without anomaly detection, then the system is simpler, but it cannot detect abnormal PCP operations in real-time
Solution Approach 1:
Machine learning models serve as intermediaries between raw sensor data and anomaly detection, learning normal operational patterns from training data and automatically identifying deviations that indicate equipment problems
Solution Approach 2:
The system transforms raw operational parameters into processed features through the machine learning pipeline, changing the representation of data to make anomalies more detectable and interpretable
4Reliability
If the event detector flags all anomalies as events, then no abnormal operations are missed, but false alarms increase and operator attention is wasted
Solution Approach 1:
The system applies partial detection by focusing on the most significant anomalies that exceed configured thresholds, rather than attempting to detect and report every minor deviation, thereby reducing false alarms while maintaining detection of critical issues
Data Source
AI summary
Systems/methods for real-time monitoring and control of a well site provide an event monitor and detector for progressing cavity pump (PCP) operations at the well site. The event monitor and detector uses machine learning (ML) based anomaly detection to detect operations that fall outside normal PCP operating space. The event monitor and detector then computes novelty scores for the anomalies and checks whether the novelty scores exceed a threshold novelty score. If the number of novelties detected within a given detection window exceeds a minimum threshold count, then the event monitor and detector flags an “event” and automatically responds accordingly. The event monitor and detector also provides an explanation with the alerts that quantifies the extent to which various PCP parameters contributed to the event. The event monitor and detector further performs drift detection to determine whether an event may be due to operator-initiated adjustments to PCP parameters.