Virtual Quality Inspection With Self-Retraining Fault Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In industrial automation, existing data classification models based on artificial intelligence struggle to efficiently adapt to new fault categories identified during production, especially in flexible production environments, where manual intervention is often required for retraining.
Innovation Solution
A method and device for virtually inspecting product quality using classifier models that dynamically retrain without human intervention by storing new fault category information in an extension buffer and using it to update the models automatically when a predefined threshold of samples is reached.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If classifier models are trained once using historic data and then deployed, then the initial classification accuracy is achieved, but the models cannot adapt to new fault categories identified during production
Solution Approach 1:
The classifier model transitions from a static, one-time training approach to a dynamic, continuous learning system. The model automatically retrains itself during production using new data from the extension buffer, enabling it to adapt to new fault categories while maintaining operational reliability.
Solution Approach 2:
The classifier model performs self-updating through automatic retraining without human intervention. The system autonomously identifies when retraining is needed, selects relevant data from the extension buffer, and updates itself, eliminating the need for manual model maintenance while preserving accuracy and enabling adaptability.
2Reliability
If manual intervention is used to retrain classifier models for new fault categories, then model accuracy can be maintained, but production downtime and complexity increase
Solution Approach 1:
The system automates the entire retraining process, from detecting the need for updates to selecting training data and executing model retraining. This self-service mechanism eliminates manual intervention, reducing operational complexity while maintaining model accuracy through continuous adaptation.
Solution Approach 2:
The system implements a feedback loop where inspection results are continuously fed back to the classifier model. When new fault categories are detected, the model receives feedback to trigger automatic retraining, creating a closed-loop system that maintains accuracy without increasing operational complexity.
3Measurement precision
If physical inspections are performed on all work in progress pieces, then inspection thoroughness is improved, but manufacturing process time and complexity increase
Solution Approach 1:
The system creates a virtual copy of the physical inspection process through classifier models that analyze production data. This virtual inspection system provides thorough quality assessment without the time consumption and complexity of physical inspections, allowing all work in progress pieces to be inspected without slowing production.
Solution Approach 2:
The patent replaces the mechanical physical inspection system with an information-based virtual inspection system using classifier models. This substitution eliminates the need for specialized inspection equipment and manual handling, maintaining inspection thoroughness while significantly improving manufacturing throughput.
Data Source
Figure 1
Figure 2~3
AI summary
The current disclosure describes a method of virtually inspecting a quality of a product in a production environment. The method comprises receiving production information associated with the product wherein the information is indicative of an operation performed on the product in relation to a first process of the production environment, determining a label for the product using a first classifier model based on the received information associated with the product, comparing the determined label against inspection data associated with an inspection of the product, and storing the production information and the inspection data associated with the product in an extension buffer, based on the comparison of the determined label and the inspection data, for retraining the first classifier model. The above method allows for storing new fault category information which is then dynamically used to retrain the model. Accordingly, this allows for retraining of classifier models without human intervention.