Beam Pump Diagnostics via Surface Dynacard and Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Beam pumps in oil wells face inefficiency issues that are difficult to diagnose, leading to lost production, health, safety, and environmental concerns, due to the uncertainty associated with surface dynacard calculations which require detailed well and pump structure knowledge, limiting their applicability across different wells.
Innovation Solution
A method and system using machine learning algorithms to analyze sensor data from beam pump units, generating a surface dynacard to predict sources of inefficiency and identify corrective actions, thereby reducing reliance on well-specific structural information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the wave equation is used to relate surface dynacard to downhole conditions, then downhole conditions can be inferred from surface conditions, but there remains a high degree of uncertainty and the system requires detailed knowledge of well and pump structure
Solution Approach 1:
The patent replaces the traditional wave equation-based mechanical/physical model with a machine learning algorithm. Instead of using physics-based calculations that require detailed structural parameters, the system uses sensor data from load cells and position indicators to train and apply ML models that directly predict downhole conditions, eliminating the need for complex structural knowledge while maintaining or improving accuracy
Solution Approach 2:
The invention changes the approach from using fixed physical parameters and structural details in wave equation calculations to using dynamic sensor measurements (load and position data) as input parameters for machine learning algorithms. This parameter transformation allows the system to adapt to different well configurations without requiring detailed structural information for each case
2Reliability
If separate wave equation calculations are performed for each well, then downhole conditions can be analyzed, but the general applicability of the system is limited between different wells
Solution Approach 1:
The patent creates a universal machine learning framework that can be applied across multiple wells and pump types. The system uses standardized sensor inputs (load and position data from surface equipment) that are compatible with various well configurations, allowing the same ML model architecture to serve multiple functions and applications across different oil wells without requiring well-specific customization
Solution Approach 2:
The system performs preliminary training of machine learning algorithms using historical sensor data and known well conditions before deployment. This pre-training phase allows the model to learn general patterns and relationships that apply across multiple wells, enabling the system to provide reliable predictions for new wells with minimal additional calibration or customization
Data Source
AI summary
A method for detecting operational issues in a beam pump unit includes receiving sensor data representing a position of and a load on the beam pump unit, using a sensor coupled to the beam pump unit, generating a surface dynacard based on the sensor data, predicting a source of inefficiency in the beam pump unit based at least in part on the surface dynacard using a machine learning algorithm, and identifying one or more corrective actions to take to address the source of inefficiency.