Fugitive Gas Quantification With Quality-Filtered AI Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hyperspectral imaging systems produce a high percentage of low-quality gas column density maps, making it challenging to accurately train artificial intelligence systems for quantifying fugitive gas leaks.
Innovation Solution
Implement a filtering mechanism to remove low-quality gas column density maps based on predefined quality metrics, such as the number of discrete plumes and plume size, before inputting them into an AI system for training a leak rate predictive model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If hyperspectral imaging systems are used to detect gas leaks, then gas leak detection capability is improved, but the percentage of low-quality gas column density maps increases
Solution Approach 1:
The patent applies preliminary action by implementing a filtering mechanism that evaluates gas column density maps against predefined quality metrics before they are used for AI training. This preliminary quality assessment ensures that only high-quality maps with appropriate plume characteristics are selected, preventing low-quality data from degrading the training process.
Solution Approach 2:
The patent applies local quality by establishing specific quality metrics for different aspects of gas column density maps, such as plume detection quality, signal-to-noise ratio, and spatial resolution. Each map is evaluated locally against these metrics to determine its suitability for training, allowing selective use of high-quality regions or maps while excluding problematic ones.
2Quantity of substance
If all gas column density maps are used for AI training, then training data quantity is improved, but training accuracy deteriorates due to low-quality maps
Solution Approach 1:
The patent applies parameter changes by dynamically adjusting the quality thresholds and filtering criteria based on the specific characteristics of the gas column density maps and the AI model requirements. This allows optimization of the balance between data quantity and quality, ensuring sufficient training data while maintaining high prediction accuracy through adaptive parameter selection.
Solution Approach 2:
The patent applies discarding and recovering by systematically filtering out low-quality gas column density maps that do not meet predefined quality metrics. This discarding process removes problematic data that would harm training accuracy, while the recovered high-quality maps form a refined dataset optimized for accurate leak rate prediction.
3Reliability
If a filtering mechanism is implemented to remove low-quality maps, then data quality is improved, but processing complexity increases
Solution Approach 1:
The patent applies parameter changes by adjusting the stringency of quality metrics and filtering thresholds to balance data quality improvement with processing complexity. By optimizing these parameters, the system achieves sufficient data quality enhancement without implementing overly complex filtering mechanisms that would unnecessarily increase processing burden.
Data Source
AI summary
Methods, apparatuses, and computer program products for fugitive gas quantification are provided. For example, a computer-implemented method may include obtaining gas column density maps associated with one or more gas leak tests, each with an associated distance and leak rate; if the density maps do not meet a predefined quality threshold, filtering the density maps to remove any that do not meet predefined quality metrics and generating a leak rate predictive model based on providing some or all of any remaining density maps that meet the predefined quality metrics, associated distances from the gas leak imaging device to the gas leak tests, and associated leak rates to an artificial intelligence algorithm; and if the density maps meet the predefined quality threshold, generating a leak rate predictive model based on providing some or all of the density maps, associated distances, and associated leak rates to the artificial intelligence algorithm.