Automation AI Control for Detecting New Production Error States
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern automation systems face challenges in reliably detecting unknown errors in production processes due to the need for extensive training data sets, which are not readily available, especially in low-error-rate production systems or when producing small quantities of products.
Innovation Solution
A method for training an artificial neural network using a meta-learning approach with strategically selected minibatches, allowing for efficient learning and adaptation to new error states with minimal additional data, enabling the AI system to recognize previously unknown errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional AI training with extensive data sets is used, then the AI system can reliably detect known error states, but it cannot efficiently detect new unknown error states due to lack of training data
Solution Approach 1:
The system performs preliminary training with extensively classified existing data before deployment. The data is pre-classified into multiple hierarchical levels (product types, error types, error severities) so that when new errors occur, the AI can leverage this pre-organized knowledge structure to quickly adapt to new error states without requiring extensive new training data
Solution Approach 2:
The invention changes the parameter of training data organization from unstructured or简单地 categorized data to extensively classified data with multiple hierarchical dimensions. This classification parameter transformation enables the AI system to efficiently generalize from known errors to unknown errors by comparing new error patterns against the structured classification framework
2Reliability
If extensive training data sets are collected for new error states, then the AI system can be retrained to detect new errors, but this requires significant time and is not feasible for low error rate production systems
Solution Approach 1:
The system performs preliminary classification and organization of training data into a hierarchical structure before it is needed for detecting new errors. This pre-processing of data classification creates a ready-to-use framework that enables rapid adaptation to new error states without requiring time-consuming data collection and training when new errors occur in production
Solution Approach 2:
The invention applies partial training action by using a reduced set of classified data for adaptation compared to conventional full retraining. The hierarchical classification allows the AI to focus learning on relevant error patterns and relationships rather than requiring extensive comprehensive training data, thus reducing training time while maintaining detection reliability
3Reliability
If conventional AI training is used for small quantity production, then the AI system can detect errors in produced items, but insufficient training data is available to train the AI system reliably
Solution Approach 1:
The system performs preliminary classification of training data into a hierarchical structure that maximizes the information content of limited data. By organizing scarce training data into multiple classification dimensions (product types, error types, severities, contexts), the AI system can extract more learning value from each data point, enabling reliable error detection even when the total quantity of available training data is small
Solution Approach 2:
The invention changes the parameter of data utilization from treating data as simple examples to organizing data into a multi-dimensional classification framework. This parameter transformation allows the AI system to learn error patterns, relationships, and hierarchies from limited data, effectively increasing the information density and learning efficiency without requiring more training data
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
The invention relates to a control unit of an automation system, which is configured to control a plant, e.g., a production plant, using, among other things, an AI system. In one application of the control unit, it monitors production with regard to the quality of the manufactured objects, e.g., the presence of defects. The AI system is trained in advance based on a large number of known states of the objects, so that the AI system is able to be trained when new, previously unknown states occur, requiring only a small number of case studies.