Bruviti
Use case

Defect Detection from SEM/AFM Images

AI-powered microscopy image analysis that auto-analyzes 90% of SEM and AFM images with 95% defect detection accuracy, reducing processing time by 85%.

Challenge

Quality teams review thousands of SEM and AFM images per lot to find and classify manufacturing defects. Analysts jump between image repositories, LIMS, PLM, defect databases, spreadsheets, and report templates. Each batch takes 2–3 days and outcomes vary by technician, limiting trend analysis and slowing release.

+
The objective

auto-analyze ≥90% of microscopy images with 95% defect detection accuracy and reduce processing time by 85%.

Solution: How the AI layer changed the operating model

Learning and setup

Powered by Bruviti’s AI Operating Layer, the agent solution applied vision, pattern, anomaly, computer vision, and LLM integration models. Training data came from historical microscopy images, annotated defect databases, technical manuals, SOP documents, and expert labeled patterns. This enabled the AI agent to recognize and interpret SEM and AFM images, batch identifiers, equipment parameters, defect types, and surface characteristics.

Agent Training workflow diagram showing training data collection, vision tool configuration, and AI capabilities development TRAINING DATA MODELS SEM AFM Images Defect DB SOP Docs Tech Manuals Expert Labels Vision AI Pattern Model Anomaly Model LLM Tools

Workflow orchestration

The AI agent analyzes incoming lots, extracts batch identifiers, and routes work across five parallel workflows: image extraction, surface analysis, defect analysis, pattern analysis, and report generation. It navigates data lake, LIMS, PLM, and report stores, mirroring the sequence QA teams follow. Orchestration logic branches when rules apply; for example, low contrast images or conflicting features trigger expert review, while high confidence detections flow to auto classification with policy checks.

SEM/AFM Workflow Orchestration diagram WORKFLOWS SYSTEMS Quality Agent Image Extraction Surface Analysis Defect Analysis Pattern Analysis Report Gen Data Lake LIMS PLM Report Store

Execution and resolution

The AI agent executes each step: extracts images from Excel and PDF sources, aligns and preprocesses frames, detects surface lines and scratches, classifies defects into seven categories, clusters patterns across samples, and generates standardized quality reports. Responses are completed in hours, with updates posted to LIMS and PLM. Exceptions such as unreadable files, novel defect signatures, or missing metadata are routed to engineers with full context and annotated visuals.

SEM/AFM Execution and Resolution workflow diagram Batch Intake Image Prep Surface Detect Classify Defects Confidence Gate Auto Classify Expert Review Pattern Cluster Report Gen Publish Results

Results

90%Drop in analysis time
2-3 hrsPer batch processing
6 monthsROI achieved

Related Warranty & Returns Use Cases

Explore by Industry

Frequently asked questions

How is AI used to detect defects in SEM and AFM microscopy images?

AI applies computer vision and pattern recognition models to scanning electron microscope and atomic force microscope images to automatically detect surface defects such as scratches, cracks, and contamination. The system classifies defects into categories, clusters patterns across samples, and generates standardized quality reports in hours instead of days.

What are the benefits of automated microscopy image analysis for manufacturing quality?

Automated microscopy analysis eliminates manual image review bottlenecks, reduces analyst-to-analyst variability, and accelerates lot release decisions. AI processes thousands of images per batch with consistent classification criteria, enabling trend analysis across production runs and achieving defect detection accuracy of 95% or higher.

How does AI image analysis reduce quality inspection time in production?

AI automates the extraction, preprocessing, and classification of microscopy images that quality teams previously reviewed manually. By running parallel analysis workflows for surface features, defect types, and pattern clusters, the system compresses batch inspection from two to three days down to a few hours while maintaining full audit traceability.