ML Model Predicts ESP KPIs to Detect Anomalies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for monitoring the performance of electrical submersible pumps (ESPs) in harsh oil and gas fields are inefficient and often result in unused or partially used high-frequency sensor data, leading to reactive maintenance and high operational costs due to frequent failures and downtime.
Innovation Solution
A data-driven workflow utilizing machine learning (ML) models to predict key performance indicators (KPIs) for healthy ESP operation, allowing for real-time monitoring and anomaly detection by training ML models with historical data to identify deviations and alert operators before failures occur.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-frequency sensors are installed to monitor ESP performance, then measurement precision is improved, but loss of information increases because much of the sensor data remains unused or partially used
Solution Approach 1:
The system implements feedback by continuously monitoring ESP performance data and automatically adjusting operational parameters. The collected sensor data feeds into machine learning models that predict remaining useful life and generate maintenance recommendations, creating a closed-loop system where measurement information is actively utilized to improve system performance and prevent failures.
Solution Approach 2:
Machine learning models serve as intermediaries between the raw sensor data and decision-making processes. These models process and interpret the high-frequency sensor data, extracting meaningful patterns and predictions about ESP health status, thereby converting the previously unused information into actionable insights.
2Device complexity
If traditional monitoring methods are used, then device complexity is reduced, but reliability deteriorates due to reactive maintenance and frequent failures
Solution Approach 1:
The system performs preliminary actions by predicting future ESP failures before they occur. Machine learning models analyze historical and real-time data to forecast remaining useful life and potential failure modes, enabling maintenance activities to be scheduled in advance. This proactive approach prevents unexpected failures and extends ESP run life without requiring complex real-time intervention systems.
Solution Approach 2:
The monitoring system enables self-service by automatically analyzing sensor data and generating maintenance recommendations without requiring constant human intervention. The machine learning models autonomously process data, detect anomalies, and provide actionable insights, reducing the need for complex manual monitoring infrastructure while improving reliability through consistent data analysis.
3Productivity
If more comprehensive data collection and analysis are implemented, then productivity is improved through better maintenance scheduling, but device complexity increases due to the need for machine learning models and data processing infrastructure
Solution Approach 1:
The machine learning platform serves multiple functions simultaneously: it collects sensor data, processes and analyzes the information, predicts remaining useful life, generates maintenance recommendations, and evaluates anomaly residuals. This multi-functional approach consolidates what would otherwise require separate systems into a single unified platform, improving productivity without proportionally increasing complexity.
Solution Approach 2:
The system monitors and responds to changes in operational parameters by adjusting maintenance schedules and recommendations. Machine learning models continuously evaluate new data against historical patterns, adapting predictions as ESP conditions change. This dynamic parameter-based approach optimizes maintenance timing and resource allocation, improving productivity through data-driven decision-making.
Data Source
AI summary
A method for monitoring operation or status of an electrical submersible pump (ESP) is provided, which includes a) collecting historical time-series data related to ESP operation: b) extracting historical time-series data related to healthy ESP operation from the historical time-series data of a): c) extracting feature data from the historical time-series data extracted in b); d) extracting or calculating values of at least one key performance indicator (KPI) related to healthy ESP operation from the historical times-series data extracted in b): c) using the feature data of c) and the values of at least one KPI of d) to train a machine learning (ML) model to predict at least one target KPI related to healthy ESP operation given feature data as input; and f) using the ML model trained in c) to monitor operation or status of the ESP. Other aspects are described and claimed.