Reproduce “Unsupervised Fault Detection using SAM with a Moving Window Approach” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ahmed Maged, Herman Shen
An unsupervised method for industrial fault detection integrates the Segment Anything Model (SAM) with a moving window approach, adaptive clustering, and an Exponentially Weighted Moving Average (EWMA) scheme. This approach achieves superior performance on industrial datasets, including up to 1.00 AUROC for textile manufacturing fault detection with continuous monitoring.
Key insights
- Leveraging a powerful, pre-trained segmentation model like SAM in conjunction with a moving window strategy significantly enhances its precision for detecting subtle, localized industrial defects without fine-tuning.
- Clustering segmented regions based on consistency across multiple overlapping windows provides a robust, unsupervised mechanism to differentiate true defects from noise or irrelevant segmentations.
- Integrating statistical process control techniques, such as EWMA, enables the developed unsupervised method to function effectively in continuous industrial monitoring, offering sensitivity to gradual defect pattern shifts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.