Orion OS: An Agentic AI System-of-Systems Framework for Cyber-Physical Optimisation in Edge Data Centre Concept Design

From site selection → design synthesis → operational validation with evidence-backed decisions

⚡

Faster Site-to-Deployment Decisions

Reduce planning cycles from weeks to hours with automated coordination across site, design, and operations teams.

🌱

Lower Energy Cost & Carbon Footprint

AI-optimised designs target PUE 1.2 with free cooling prioritisation and intelligent load forecasting.

✓

Engineering-Compliant by Design

Every recommendation grounded in industry standards, equipment specifications, and regulatory constraints.

📊

Evidence-Based, Explainable Outputs

RAG-powered explanations cite sources, standards, and engineering rationale for every design decision.

🎯 Architect: Engineering Design Team View

Meeting Scenario

Persona: Engineering Design Team

Task: Design a 500 kW edge data centre, PUE target 1.2, N+1 redundancy, no water restriction assumed unless provided by site envelope.

User Intent

500 kW
1.2
N+1

🧠 RAG Evidence and Decision Rules

Click "Run Full Coordination Demo" above to see RAG-backed evidence for each agent decision.

System Architecture: System-of-Systems Multi-Agent Platform

🎯 Orchestrator

Manages decision flow, feedback loops, and cross-agent context sharing

↓
1

Site Agent

Find feasible grid-connected locations

Inputs

Power capacity, grid policy, network density

Outputs

Shortlisted viable sites with scores

Example: 2.5 MVA site, 35% utilisation
2

Design Agent

Synthesize compliant physical design

Inputs

Site constraints, power availability

Outputs

Cooling strategy, UPS topology, PUE target

Example: Hybrid cooling, N+1 UPS, PUE 1.28
3

Operation Agent

Validate real-world performance

Inputs

Proposed design, load forecasts

Outputs

PUE validation, capacity risk, energy forecast

Example: PUE 1.32, 96% UPS efficiency
🧠

Shared RAG Knowledge Base

Evidence, standards, and decision rules used across all agents

Contains

Engineering standards, equipment specs, best practices, diagnostic rules

Accessible By

All agents for context-aware decision support

↕ Queries & Evidence ↕

🧠 Shared RAG Knowledge Base

Engineering standards · Equipment datasheets · Best practices · Diagnostic rules · Regulatory constraints

Data Flow & RAG Knowledge Base

📍 Data Flow

  • (i) Site Agent produces machine-readable feasibility envelopes (e.g., grid headroom, electrical capacity, fibre, property/planning limits, regulatory constraints, cost bounds).
  • (ii) Design Agent converts envelopes into a constrained, multi-objective design space (e.g., UPS topology, cooling topology, redundancy allocation, supervisory control variables) and generates candidate configurations with KPIs (e.g., projected PUE, redundancy margin, thermal operating range).
  • (iii) Operations Agent validates candidates in closed loop using a Dymola-powered Cyber-Physical System (CPS) exported as a Functional Mock-up Unit (FMU) for co-simulation, integrating two-layer LSTM load forecasting with hierarchical MPC to test dynamic robustness under demand variation.
  • (iv) Finally, a RAG layer retrieves scenario-aligned standards and supplier specifications; retrieved constraints are injected into optimisation bounds and logged in the digital thread for traceable compliance evidence.

💡 Evidence Flow (RAG at Every Stage)

  • Standards: ASHRAE thermal guidelines, BS EN 50600 compliance
  • Equipment Curves: UPS efficiency maps, chiller performance data
  • Best Practices: Free cooling thresholds, N+1 redundancy rules
  • Diagnostic Rules: PUE anomaly detection, load balancing heuristics

Every recommendation is traced back to its engineering source.

Detailed Agent Exploration

For in-depth analysis and manual configuration

Step 1: Select Site

Not Started
Explore detailed site rankings with interactive map, filter by utilisation, capacity, and network density.

Step 2: Generate Design

Not Started
Review proposed cooling, UPS, and PUE targets with RAG-backed engineering justifications.

Step 3: Validate Operations

Not Started
Inspect real-time energy metrics, load forecasts, and PUE validation dashboards.

Who Uses This Platform

🏗️

Architect

What they get:

Site and system-level design, growth intent, budget and scalability.

⚡

Electrical / Mechanical Engineer

What they get:

UPS, cooling modality, heat rejection, setpoints, sizing assumptions.

🔧

Data Centre Manager

What they get:

Stability, maintenance windows, operational recommendations.

📋

Risk & Compliance Lead

What they get:

Evidence pack, thresholds, audit trail, concept freeze approval.

🟢 Live Demo Context

No workflow started yet. Click Run Full Coordination Demo above, or open Step 2: Generate Design.

🧠 RAG Knowledge Base Console

Example: "What cooling strategy is recommended for UK climate?"
Auto-filled from Live Demo Context. Edit or leave empty.
Use last run context

Querying RAG knowledge base...

Results