Hardware-to-Brain
Intelligence Architecture
A unified three-layer vertical architecture that eliminates traditional factory "data silos" and "decision delays" through seamless integration from edge sensing to AI-powered strategic decision-making.
🏗️ Three-Layer Vertical Integration
⚡ The Edge Foundation - Powered by NXP
The foundation layer leverages NXP i.MX RT1170 Dual-Core Crossover MCU with eIQ® Time Series Studio to achieve millisecond-level edge inference and anomaly detection.
🔧 Hardware Core
NXP i.MX RT1170
Dual-core heterogeneous architecture (Cortex-M7 + M4) for simultaneous sensor sampling and AI
inference
📡 Ultra-High Frequency Sampling
10kHz vibration data stream
Real-time monitoring with anomaly detection latency < 4.8ms
🤖 Edge Inference
eIQ® AutoML Pipeline
On-device FFT/RMS feature extraction with model compression and optimization
📊 Data Push-Down
90% edge preprocessing
Only critical health scores uploaded to cloud, reducing bandwidth by 95%
🗺️ The Analytics Matrix - Digital Twin & Spatial Intelligence
The analytics layer provides real-time 3D digital twin visualization, multi-floor spatial management, and comprehensive fleet health monitoring through an intuitive dashboard interface.
🎮 3D Digital Twin
Three.js / WebGL
Real-time rendering with data-driven materials. Machine colors dynamically change based on
Health Score
🏢 Multi-Floor Management
1F / 2F / 3F Navigation
Independent floor mapping with drag-and-drop layout editing for Superusers
📈 Fleet Health Score
Multi-dimensional fusion
Real-time scoring (0-100) with data integrity indicators
🔄 Closed-Loop Optimization
AI-driven parameter tuning
Gemini 3.0 analyzes long-term failure trends and suggests eIQ threshold adjustments
🧠 The Agentic Matrix - Powered by 3rd Party LLM Model
The decision layer features an 8-Agent collaborative matrix powered by 3rd Party LLM models, providing knowledge-driven reasoning and automated decision-making with human-in-the-loop approval.
📥 Ingestion Agent
IoT data ingestion via MQTT / StreamBuffer
🔧 Preprocessing Agent
Signal cleaning through NXP eIQ Pipeline
🔍 Diagnostic Agent
Fault diagnosis using eIQ Classifiers
💡 Lydia (Expert)
On-premises Model - Maintenance SOP recommendations
🌟 Gemini 3 (Core)
Google Gemini 3.0 Pro - Strategic long-term reasoning
✅ Quality Agent
Yield prediction and quality assurance
💰 Cost Agent
ROI analysis and cost optimization
👔 Leader Agent
HITL approval and work order dispatch
🔗 Live Expert API Integration
The system seamlessly integrates with external AI experts through secure API connections:
- On-premises Model: Direct Line API for SOP query and maintenance guidance
- Google Gemini 3.0 Pro: Spectrum UI with iridescent breathing light interface for high-level strategic reasoning
- Data Push-Down: Structured JSON data directly injected as Prompt Context for Agentic RAG processing
🚀 System Highlights
🌐 Internationalization
Automatic language detection (ZH/EN) with context-aware terminology switching
👥 Human-in-the-Loop (HITL)
Engineers can supplement field observations anytime. AI consensus requires human approval before ERP dispatch
🔐 Role-Based Access Control
Four-tier permissions: Admin / Superuser / Normal User / Visitor with JWT authentication
📚 Knowledge Base RAG
Structured knowledge entities including factory SOPs, maintenance history, and ISO standards
Experience the Future of
Intelligent Manufacturing
From edge sensing to AI-powered strategic decisions, our three-layer architecture delivers unparalleled predictive maintenance capabilities for semiconductor and high-tech manufacturing.