Drilling Rig State Detection Using Video Instead of Sensors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current rig state detection methods rely on expensive and prone-to-failure sensor data, necessitating a more reliable and cost-effective alternative.
Innovation Solution
Implement video classification techniques using machine learning models trained on optical images and video data from cameras positioned around the rig, employing convolutional neural networks to infer rig states, and an annotation pipeline to process and annotate video data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sensor data is used for rig state detection, then detection accuracy can be achieved, but the system becomes expensive and prone to failure
Solution Approach 1:
The patent replaces the mechanical sensor-based detection system with a computer vision system using cameras and machine learning models. Video cameras capture rig operations, and trained neural networks classify rig states from visual data, eliminating the need for physical sensors on the rig while maintaining detection capability
Solution Approach 2:
The patent creates a virtual model of the rig state by capturing visual information through cameras and processing it through machine learning models. Instead of directly measuring physical parameters with sensors, the system creates a digital representation of rig states from video imagery, enabling detection without physical contact or sensor installation
2Productivity
If sensor data is used for rig state detection, then real-time detection is possible, but maintenance costs increase
Solution Approach 1:
The machine learning models are trained once on labeled video data and then autonomously perform rig state classification without requiring ongoing maintenance. The system serves itself by continuously processing video streams and generating classifications, eliminating the need for sensor calibration, replacement, or maintenance that would be required for physical sensor systems
3Ease of manufacture
If video data is used instead of sensor data, then cost and maintenance are reduced, but detection accuracy may be compromised
Solution Approach 1:
The system performs preliminary training of machine learning models using extensively labeled video data before deployment. This pre-training phase creates robust models that have already learned to accurately classify rig states from visual features, ensuring high detection accuracy is achieved before the system begins operational use
Solution Approach 2:
The patent transforms the detection problem by changing the input parameters from physical sensor measurements to visual features extracted from video frames. By adjusting the model architecture, training data composition, and feature extraction parameters, the system optimizes accuracy for video-based detection while maintaining cost-effectiveness
Data Source
Figure 1
Figure 2A~2B
Figure 3~4
AI summary
A method includes receiving training images representing a portion of a drilling rig over a first period of time, associating individual training images of the training images with times at which the individual training images were captured, determining a rig state at each of the times, classifying the individual training images based on the rig state at each of the times, training a machine learning model to identify rig state based on the classified training images, receiving additional images representing the portion of the drilling rig over a second period of time, and determining one or more rig states of the drilling rig during the second period of time using the machine learning model based on the additional images.