GAM Training With Embeddings for Manufacturing Defect Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing quality assurance methods in manufacturing fail to provide deeper insights into the specific causes and types of defects, leading to inefficient production improvements and unnecessary rejection of components.
Innovation Solution
A two-stage process using a generalized additive model (GAM) and an Explainable Boosting Machine (EBM) to analyze sample characteristics, followed by embedding vectors to train a defect type clustering model, enabling deeper defect analysis and automated classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a simple OK/NOK classification system is used, then the distinction between acceptable and defective parts is effective, but deeper insight into specific defect causes and types is not provided
Solution Approach 1:
The patent segments the defect analysis process into two distinct stages: first, a GAM model segments samples into OK/NOK categories, and second, a clustering model segments the NOK samples into different defect type clusters. This segmentation allows the system to maintain simple binary classification while also providing detailed defect type information through subsequent clustering of the rejected samples.
Solution Approach 2:
The patent introduces embedding vectors as an intermediary representation between the GAM classification output and the final defect type clustering. These embedding vectors capture the contribution of each input feature to the NOK classification, serving as a mediator that preserves defect characteristic information while enabling unsupervised clustering of defect types without requiring manual labeling.
2Loss of information
If manual labeling by defect type is performed, then detailed analysis based on specific defect types is enabled, but time and cost are increased
Solution Approach 1:
The patent implements self-service by using unsupervised clustering algorithms that automatically group defective samples into defect type classes based on their embedding vectors, without requiring manual human labeling. The system serves itself by extracting meaningful defect type information from the data through the GAM-derived embeddings and applying clustering techniques, thereby eliminating the time-consuming manual labeling process while preserving detailed defect type information.
3Ease of operation
If glass-box models like GAM are used, then transparency and interpretability of decision-making are improved, but the ability to perform unsupervised defect clustering is limited
Solution Approach 1:
The patent merges the strengths of glass-box interpretability and unsupervised clustering capability by combining GAM models with embedding-based clustering. The GAM provides transparent, interpretable feature contributions to defect detection, while the extracted embedding vectors from the GAM are then used as input to unsupervised clustering algorithms. This combination allows the system to maintain model interpretability while gaining the versatility to perform automated defect type clustering.
Data Source
Figure 1~2
AI summary
The invention relates to a method and a device (100) for training a generalized additive model (GAM) for detecting defective samples in a manufacturing environment and for training a defect type clustering model based on the defective samples. Furthermore, the invention relates to a method for detecting defective samples in a manufacturing environment and/or for clustering the samples identified as defective into multiple defect type classes.