Industrial AI Target Data Generation From Process State Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The laborious and resource-intensive process of data collection and annotation for machine learning in industrial settings, particularly in identifying relevant data for training and evaluating models, is inefficient due to human subjectivity and high data volumes from sources like sensors and image capturing devices, leading to suboptimal model performance.
Innovation Solution
A computer-implemented method that identifies target states in industrial processes using monitoring data from various sources, generating target data based on detected states, which can include image or video data, to streamline data selection and annotation for machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human annotators manually analyze and annotate the full collected dataset, then data annotation can be completed, but the process becomes very time-consuming and inefficient
Solution Approach 1:
The system extracts only the relevant subset of data that represents target states, rather than annotating the full dataset. This is achieved by automatically identifying and selecting data portions that contain meaningful information for machine learning models, significantly reducing annotation time while maintaining quality.
Solution Approach 2:
The system performs self-annotation by automatically identifying target states and generating annotations without human intervention. The automated annotation process analyzes monitoring data, detects target states, and creates annotations independently, eliminating the time-consuming manual annotation process while maintaining consistent quality.
2Reliability
If the full collected dataset is analyzed by annotators, then comprehensive data coverage is achieved, but the analysis becomes very time-consuming and inefficient
Solution Approach 1:
The system extracts only the relevant subset of data that represents target states, rather than processing the full dataset. This extraction process identifies and isolates meaningful data portions, achieving comprehensive coverage of important states while dramatically improving annotation efficiency.
Solution Approach 2:
Instead of annotating the complete dataset, the system performs partial annotation on only the relevant subsets containing target states. This partial action approach focuses computational and human resources on the most valuable data portions, improving productivity without sacrificing reliability.
3Ease of operation
If human annotators identify relevant data subsets, then data selection can be performed, but quality and relevance may vary due to human subjectivity
Solution Approach 1:
The system performs self-annotation by automatically identifying target states and generating annotations without human intervention. This automated approach eliminates human subjectivity and ensures consistent, objective data relevance assessment across all annotations.
Solution Approach 2:
The system uses feedback from monitoring data analysis to automatically adjust and refine data selection criteria. By continuously learning from the data patterns and target state identifications, the system maintains high consistency in data relevance without human subjectivity.
4Reliability
If monitoring data from sensors and image capturing devices is collected, then comprehensive process information is obtained, but data volume becomes high posing handling challenges
Solution Approach 1:
The system extracts only the relevant subsets of monitoring data that represent target states, rather than storing and processing the entire high-volume dataset. This extraction significantly reduces data volume for handling and storage while maintaining the accuracy needed for process monitoring and machine learning applications.
Data Source
AI summary
A computer implemented method for determining target data for artificial intelligence is disclosed. In one aspect, the method may include identifying an indication of a target state of a first industrial process or asset, and/or of a second industrial process or asset operatively associated with the first industrial process or asset, and determining target data, based on the identified indication of the target state. The identification of the indication of the target state may be based on first monitoring data indicative of data output from first monitoring data source(s) associated with the first industrial process or asset or with the second industrial process or asset. The target data may be determined from second monitoring data indicative of data output from second monitoring data source(s) associated with the first industrial process or asset.