Explosion Prediction Using Sensor-Constructed Risk Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems struggle to accurately predict explosions by effectively integrating gas and flame sensor data due to issues like physical blockages and inaccuracies in image capture, leading to unreliable explosion predictions.
Innovation Solution
A computer-implemented method that constructs a data-constructed image using gas and flame sensor data, assigning pixel values to different channels, and applies a specially trained machine learning model to determine explosion contribution levels, generating accurate explosion prediction data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image capture methods are used to detect explosion conditions, then visual information can be obtained, but physical blockages and inaccuracies reduce prediction reliability
Solution Approach 1:
The patent replaces optical/image capture systems with a data-constructed image approach that uses sensor data (gas concentration, flame detection, temperature) to create a synthetic representation of the environment. This substitution eliminates the problems of physical blockages and optical inaccuracies by using direct sensor measurements that are not affected by line-of-sight issues.
Solution Approach 2:
The patent introduces a data-constructed image as an intermediary between raw sensor data and explosion prediction. This intermediary synthesizes multiple sensor inputs (gas sensors, flame sensors, temperature sensors) into a unified representation that the machine learning model can process, improving both accuracy and reliability by combining multiple measurement sources.
2Measurement precision
If multiple sensor types are integrated for comprehensive detection, then prediction accuracy improves, but system complexity increases
Solution Approach 1:
The patent merges multiple sensor types (gas sensors, flame sensors, temperature sensors) into a unified data-constructed image representation. By combining these different data sources into a single synthesized image structure, the system achieves comprehensive detection accuracy while managing complexity through a unified processing approach rather than separate analysis streams.
Solution Approach 2:
The data-constructed image serves as a universal data structure that can accommodate multiple sensor types and their respective measurements. This multi-functional representation allows the same machine learning model to process diverse sensor inputs (gas concentrations, flame presence, temperature) through a single unified interface, reducing overall system complexity.
3Device complexity
If existing sensors are utilized instead of adding new hardware, then cost and hardware complexity are reduced, but data integration challenges arise
Solution Approach 1:
The data-constructed image creates a universal data structure that can integrate measurements from various existing sensor types. This unified representation allows the system to leverage existing gas sensors, flame sensors, and temperature sensors without requiring new specialized hardware, while the standardized image format simplifies the integration of heterogeneous data sources.
Solution Approach 2:
The patent transforms sensor readings (gas concentration values, flame intensity, temperature measurements) into pixel values within a standardized image data structure. This parameter transformation converts diverse sensor outputs into a common format that can be easily processed by the machine learning model, reducing integration difficulty.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Embodiments utilize captured data, such as gas data and/or flame/heat data, from sensors in an environment to generate a data-constructed image for use in predicting explosion likelihood within an environment. Some embodiments utilize gas and flame data to generate the data-constructed image that is processable via one or more model(s) to determine whether the environment includes one or more sub-regions at risk of explosion. Some embodiments receive a plurality of gas sensor data and a plurality of flame sensor data, generate a data-constructed image including a plurality of channels based at least in part on such data, and generate explosion prediction data by applying at least a portion of the data-constructed image to a prediction model.