Facility State Detection Using Grouped Signal Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As production facilities scale up and the number of signals increases, existing anomaly detection technologies face challenges with longer learning times and larger data requirements due to complex signal relationships, especially when processing multiple workpieces asynchronously.
Innovation Solution
A state detection system that collects and divides signal data chronologically into groups, allowing for machine learning on each group to generate a normal model, enabling efficient detection of facility states with reduced learning time and data needs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the scale of the production facility increases and the number of signals increases, then the facility can process more workpieces in parallel, but the relationships between signals become complex, increasing the required learning time and required amount of learning data
Solution Approach 1:
The patent divides the facility signals into multiple groups based on their relationships and characteristics. Instead of learning all signals simultaneously in a single large model, the system creates multiple smaller learned models, each handling a specific group of signals. This segmentation reduces the complexity of learning for each model, thereby reducing the overall learning time while maintaining the capability to handle parallel workpiece processing.
2Productivity
If the scale of the production facility increases and the number of signals increases, then the facility can process more workpieces in parallel, but the relationships between signals become complex, increasing the required amount of learning data
Solution Approach 1:
By segmenting the signals into multiple groups and creating separate learned models for each group, the patent reduces the amount of learning data required for each individual model. Each smaller model only needs to learn the relationships within its specific signal group, rather than learning all signal relationships in the entire facility, thus reducing the total learning data requirement while maintaining high productivity.
3Reliability
If a single learned model is used to cover the behavior of the entire facility, then the model must account for all signal relationships, but this increases the learning time and learning data requirements
Solution Approach 1:
The patent segments the facility into multiple signal groups, each with its own learned model. This approach maintains comprehensive facility state detection reliability by covering all signals through the combination of multiple specialized models, while reducing learning time by allowing each model to learn only its specific signal group independently and in parallel.
Solution Approach 2:
The patent combines multiple specialized learned models into an integrated system that collectively monitors the entire facility. Each model handles its specific signal group, and their results are integrated to provide comprehensive facility state detection, achieving both reliability and efficiency.
Data Source
AI summary
A collection unit (111) collects collected data of a plurality of signals in a facility in chronological order. A division unit (112) divides the collected data in chronological order into a plurality of groups so as to generate collected divided data for each group. A learning unit (113) performs machine learning for each group, using the collected divided data as learning data, so as to generate a normal model, which is a learned model, for each group. A state detection unit detects a state of the facility, using the normal model of each group.