Drilling Event Classification via Supervised Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional alarm systems in drilling operations are ineffective in detecting well control events such as kicks and losses due to high rates of false alarms, loss of event signatures in normal data variance, and difficulty in taking alarms seriously due to frequent false alerts.
Innovation Solution
A method and system that utilize machine learning to classify drilling events by determining relevant drilling attributes, calculating correlation values, selecting a set of attributes, generating a labelled dataset, and training a supervised machine learning model to accurately categorize drilling events into predefined categories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional alarm systems use simple thresholds for mud volume and flow rate changes, then the system is easy to operate, but it generates many false alarms and loses event signatures in normal data variance
Solution Approach 1:
The patent transforms the alarm system from using simple fixed thresholds to using dynamic, context-aware probability thresholds. The system calculates a probability value based on multiple drilling parameters and their relationships, allowing the alarm threshold to adapt to normal operational variations rather than using static cutoffs that generate false alarms.
Solution Approach 2:
The patent introduces an intermediary probability calculation layer between the raw drilling data and the alarm output. This intermediary layer processes multiple parameters (mud volume, flow rate, pump rate, etc.) and their correlations to produce a probability value, which then determines whether an alarm should trigger, filtering out false alarms while maintaining reliability.
2Device complexity
If conventional alarm systems use traditional alarm sounds, then the system is simple, but drilling team members find it difficult to take alarms seriously due to false alarms
Solution Approach 1:
The patent implements feedback by continuously monitoring drilling parameters and updating the probability calculation in real-time. The system provides feedback to operators through the probability value display, allowing them to understand the system's confidence level in detecting actual events versus normal variations, thereby improving their response to alarms.
3Reliability
If the system monitors multiple drilling attributes to improve detection accuracy, then event detection reliability improves, but the complexity of the system increases
Solution Approach 1:
The patent segments the complex monitoring task into distinct functional components: parameter collection module, correlation analysis module, probability calculation module, and alarm generation module. This segmentation allows the system to handle multiple drilling attributes systematically while maintaining manageable complexity through modular architecture.
Solution Approach 2:
The patent applies partial action by selectively monitoring and analyzing only the most relevant drilling parameters for well control events, rather than attempting to monitor all possible parameters. The system identifies and processes key attributes (mud volume, flow rate, pump rate, etc.) that have the highest correlation with well control events, reducing complexity while maintaining high detection reliability.
Data Source
AI summary
The invention relates to method and system for classifying events (well control events) during drilling operations. The method includes determining drilling attributes corresponding to volve drilling data in a predefined format and associated with one or more wells; determining a correlation value between each two attributes of the drilling attributes associated with the volve drilling data; selecting a set of drilling attributes from the drilling attributes based the determined correlation value; generating a labelled dataset corresponding to the volve drilling data based on the set of drilling attributes by determining a value of one or more additional drilling attributes associated with the volve drilling data based on the set of drilling attributes; and training a supervised Machine Learning (ML) model based on the labelled dataset and real-time drilling data for classifying each of the one or more drillings events in one of a set of pre-defined categories.