AI Vision Inspection & Monitoring System

AI Vision IV4Next.jsReactPLCDatabase Integration

projectDetail.role

Full Stack Industrial Developer

projectDetail.year

2025

projectDetail.duration

4 Months

projectDetail.type

projectDetail.industrialAutomation

projectDetail.theChallenge

The existing rule-based vision inspection system caused frequent micro line stops, averaging 230 stops and 16.5 minutes of downtime per day. It suffered from low inspection stability, high false NG rates, and a high dependency on vendor services for any model changes, which resulted in continuous extra costs and production instability.

projectDetail.theSolution

Replaced the legacy system with a new AI-based Vision IV4 system integrated with a custom Next.js Real-Time Monitoring Dashboard. This allows flexible, in-house AI model training and provides instant visual feedback, historical image galleries, and live yield rate tracking directly from the production line.

projectDetail.transformationEvidence

projects.before

✓ projects.after

Inspection Stability

← drag to compare →

projectDetail.systemDemo

projectDetail.systemArchitecture

projects.systemArchitectureLive

Camera

AI Vision IV4

Controller

PLC

Database

PostgreSQL

Dashboard

Next.js Web App

projects.latency: 12ms

projects.powerStable

projectDetail.lessonsLearned

- Data Verification vs Dashboard Reality: During live monitoring, the dashboard displayed 4 NGs out of 1640 scans (Yield Rate 99.8 percent). Upon cross-checking with the activity history, I found that only 2 were actual camera-detected NGs, while the other 2 were manually forced NGs made by QC for system testing. This taught me the importance of building detailed activity logs to verify actual production quality versus system test data. - Eliminating Vendor Bottlenecks: The shift from a closed, rule-based system to an AI learning model enabled our internal team to handle model changes and expansions independently, saving significant time and recurring vendor service fees.

projectDetail.businessImpact

Line Stops Reduced 89.6 Percent | Recovered 65 Hours/Year

projects.detailHeading

Based on the Vision System Improvement Report, this project completely overhauled the quality control process for washing machine component assembly. System Architecture: 1. Inspection Layer: AI Vision IV4 smart cameras capture and analyze components (checking wires, screws, and assembly points) using advanced machine learning models. 2. Data Processing: The camera communicates directly with the PLC for reject logic and sends scan data to the centralized database. 3. Real-Time Dashboard: A Next.js application visualizes live data, showing total production (over 24,000 items), yield rates, and a detailed image gallery of every scanned part with exact timestamp tracking. Key Improvements and Business Value: - Massive Downtime Reduction: Slashed line stop frequency by 89.6 percent (from 230 to 24 times per day), reducing daily downtime from 16.5 minutes to just 1.7 minutes. - Production Time Recovery: The efficiency gain translates to approximately 65 hours of recovered production capacity per year. - Strategic Independence: Transitioned from a closed, vendor-dependent system to a highly flexible, in-house manageable solution, perfectly aligning with the smart factory roadmap.

projects.technologies

AI Vision IV4
Next.js
React
PLC
Database Integration

projects.systemGallery

projects.totalAssets: 12

projects.needSystem

projects.ctaText

projects.discussProject

projectDetail.nextProject

DJ MES: Smart Factory Operations Control Center

Next.js • TypeScript • React Native