OT Digital Twin (OTDT)

Agentic AI Operating Layer for Geothermal Critical Infrastructure

⚠️ SYNTHETIC DATA: All sensor readings and AI predictions are generated from simulation models. Not real plant data.

Real-time industrial digital twin for geothermal plant operations — Monte Carlo failure prediction, LSTM anomaly detection, and live Unity XR visualisation. Built on IBM Maximo MAS 9.1 + watsonx.ai + Red Hat OpenShift.

East Africa Agentic AI Workshop 2026 — IEEE IES East Africa Industrial Innovation Summit

---

Architecture Overview

Physical Sensors → Watson IoT → Kafka → API (FastAPI)
                                          ├── TimescaleDB (time-series storage)
                                          ├── Redis (real-time state cache)
                                          ├── Anomaly Detector → Monte Carlo Engine
                                          │                       └── Maximo Work Orders
                                          └── WebSocket → React Dashboard / Unity XR

Five AI Agents

Agent	Description	Model
Failure Predictor	LSTM model trained on 5-year sensor history. Predicts failure probability over 30/60/90-day windows	LSTM (3-layer, 64 units)
Monte Carlo Sim Engine	10,000 Weibull-distributed failure scenarios against live sensor readings	NumPy + SciPy
What-If Analyst	Interactive maintenance deferral slider — quantifies risk of deferral decisions in real time	Monte Carlo + cost model
Anomaly AI	CNN visual anomaly detection on 3D digital twin screenshots	ResNet-18 fine-tuned
Maintenance Scheduler	Converts failure curves into optimised inspection schedules	Priority queue + LP

Quick Start

Docker Compose (recommended)

docker-compose up --build

Services:

• API: http://localhost:8000 (FastAPI + Swagger at /docs)
• Dashboard: http://localhost:3000 (React)
• Kafka: localhost:9092 (internal), localhost:29092 (external)
• TimescaleDB: localhost:5432
• Redis: localhost:6379

Local Development

# API
pip install -r requirements.txt
uvicorn api.main:app --host 0.0.0.0 --port 8000 --reload

# Sensor Simulator
cd sensor_simulator
python kafka_publisher.py --demo

# Frontend
cd frontend/react-dashboard
npm install && npm start

Project Structure

OTDT/
├── api/                    # FastAPI application
│   ├── main.py             # App entry point, WebSocket, Kafka consumer
│   ├── routers/            # Route handlers (whatif, anomaly, predict, assets, etc.)
│   ├── anomaly/            # Rule-based anomaly detector
│   ├── db/                 # TimescaleDB + Redis clients
│   └── integrations/       # Kafka consumer, Maximo client
├── monte_carlo/            # Weibull-based MC simulation engine
│   └── engine.py           # run_simulation() + whatif_simulation()
├── ml/                     # Machine learning models
│   ├── lstm/               # LSTM failure predictor (Step 4)
│   └── cnn_anomaly/        # CNN visual anomaly detector (Step 7)
├── sensor_simulator/       # Geothermal sensor telemetry simulator
├── scheduler/              # Maintenance schedule optimizer (Step 8)
├── maximo/                 # IBM Maximo MAS integration layer
├── frontend/react-dashboard/  # React monitoring dashboard
├── unity/
│   ├── GDC_Plant_Twin/     # Unity XR 3D digital twin
│   └── ThreeJS_Viewer/     # Browser-based 3D fallback
├── datasets/               # GDC Kenya synthetic datasets (Excel)
├── docker-compose.yml      # Full stack orchestration
└── requirements.txt        # Python dependencies

Demo Scenarios

1. Failure Prediction: WP-07 shows 34% failure probability within 30 days — current calendar schedule wouldn't inspect for 45 days
2. What-If Analysis: Defer maintenance slider 0→45 days, probability rises 34%→68%, expected cost USD 122,400
3. Optimised Schedule: 90-day schedule highlighting 3 unplanned high-risk assets

Key Endpoints

Method	Path	Description
GET	/health	Service health check
GET	/api/assets	All 50 GDC geothermal assets
GET	/api/twins/{asset_id}/sensors/latest	Latest sensor readings
POST	/api/monte-carlo/simulate	Run 10k Monte Carlo simulation
POST	/whatif/simulate-days	What-If deferral analysis
GET	/api/anomaly/status	CNN anomaly scores (all assets)
POST	/api/predict/failure	LSTM failure probability
WS	/twin/stream	Real-time WebSocket feed

Technology Stack

IBM: Maximo MAS 9.1 APM · Watson IoT · Watson Studio ML · watsonx.ai · Event Streams (Kafka) · Red Hat OpenShift

Core: Python 3.11 · FastAPI · NumPy · SciPy · TensorFlow · Kafka · TimescaleDB · Redis

Visualisation: React · Unity Pro 2023.2 LTS · Three.js

Target Client

GDC Kenya (geothermal) — 180 wells, 50 modelled assets. ROI: 650%+ (preventing 2 unplanned pump failures/year saves USD 360,000 vs. USD 48,000 platform cost).

---

i3 Technologies Ltd | IBM Silver Partner CEID 7sq30 | Philip Mukiti | 2026