⚙️ PLCpilot — Self-Evolving PLC Automation at the Edge

GenAI-powered industrial automation that detects anomalies, generates PLC logic, and deploys updates autonomously — with human-in-the-loop oversight.

---

🚀 Overview

PLCpilot is an AI-driven edge automation platform that bridges the gap between natural language intent and deployable PLC code. It uses a 7-agent pipeline to:

1. Detect sensor anomalies in real time
2. Identify root causes automatically
3. Generate OpenPLC-compatible Structured Text (ST) code via LLaMA3 (Ollama)
4. Deploy updates to OpenPLC without manual intervention
5. Notify human operators and learn from outcomes

---

🏗️ Architecture

Sensor Data (Node-RED/MQTT)
         |
         v
+---------------------+
|  Anomaly Detection  |  <-- Detects drift, out-of-range, missing pulses
+---------------------+
         |
         v
+---------------------+
|  Root Cause Agent   |  <-- Maps anomaly to cause category
+---------------------+
         |
         v
+---------------------+
|  Prompt Generation  |  <-- Converts cause to GenAI prompt
+---------------------+
         |
         v
+---------------------+
|  GenAI Code Agent   |  <-- Ollama LLaMA3 -> Structured Text (.st)
+---------------------+
         |
         v
+---------------------+
|  Deployment Agent   |  <-- Pushes to OpenPLC, monitors compile
+---------------------+
         |
         v
+---------------------+
|  Human Notification |  <-- Email/MQTT alert, approval request
+---------------------+
         |
         v
+---------------------+
|   Learning Agent    |  <-- Logs outcomes, improves future prompts
+---------------------+

---

📁 Repository Structure

plcpilot/
├── app/
│   ├── app.py                       # Flask app: starts/stops MQTT simulation via web UI
│   ├── streamlit_app.py             # Streamlit GenAI PLC code generator
│   └── templates/
│       └── index.html               # Simulator control UI (Start/Stop buttons)
├── plc_programs/
│   ├── temp_fan_control.st          # Temperature-based fan/alarm control
│   ├── watchdog_press.st            # Press machine watchdog timer (10s)
│   ├── pump_alternation.st          # Pump A/B 12-hour alternation
│   └── sensor_drift_robotic_arm.st  # Robotic arm drift detection
├── nodered_flows/
│   ├── plcpilot_flow.json          # Node-RED Modbus simulation flow
│   └── nodered_function.js          # Function node: 10 machines × 12 tags → InfluxDB + MQTT
├── agents/
│   └── pipeline.py                  # 7-agent pipeline description & runner
├── mosquitto/
│   └── config/
│       └── mosquitto.conf           # MQTT broker config (port 1883, persistence enabled)
├── docs/
│   └── ARCHITECTURE.md              # Modbus mapping & Node-RED setup reference
├── Dockerfile                       # Flask simulator container (python:3.11-slim)
├── docker-compose.yml               # Full stack: Mosquitto + Flask on iot_net
├── requirements.txt                 # Python dependencies
└── .gitignore

---

🛠️ Tech Stack

Component	Technology
PLC Runtime	OpenPLC (open-source)
PLC Language	Structured Text (IEC 61131-3)
GenAI Engine	Ollama + LLaMA3 (local inference)
Web UI	Streamlit
Sensor Simulator	Flask + Paho MQTT
Message Broker	Eclipse Mosquitto
Field Bus	Modbus TCP
Flow Automation	Node-RED
Containers	Docker + Docker Compose

---

⚡ Quick Start

Prerequisites

• Docker & Docker Compose
• OpenPLC Runtime installed locally
• Ollama with LLaMA3 pulled: ollama pull llama3
• Node-RED with node-red-contrib-modbus installed

1. Clone and start services

git clone https://github.com/YOUR_USERNAME/plcpilot.git
cd plcpilot
docker-compose up -d

2. Run the Streamlit code generator

pip install -r requirements.txt
streamlit run app/streamlit_app.py

3. Import Node-RED flow

• Open Node-RED at http://localhost:1880
• Import nodered_flows/plcpilot_flow.json
• Deploy the flow

4. Deploy a PLC program

• Open OpenPLC at http://localhost:8080
• Upload any .st file from plc_programs/
• Start the runtime

5. Test

• Node-RED injects temperature values via Modbus to %MW0 (register 1024)
• OpenPLC evaluates the logic and sets coil outputs
• Read coils 0 and 1 in Node-RED to observe fan/alarm state

---

🧪 Demo Use Cases

Use Case	File	Modbus Register
Temperature Fan Control	temp_fan_control.st	Write to HR 1024 (%MW0)
Press Machine Watchdog	watchdog_press.st	Toggle Coil 0 (%IX0.0)
Pump A/B Alternation	pump_alternation.st	Read Coils 0,1
Robotic Arm Drift	sensor_drift_robotic_arm.st	Write HR 1024, 1025

---

🤖 AI Agent Pipeline

See agents/pipeline.py for the full 7-agent description.

The agents are designed to be modular — each can be swapped for a production service (e.g., replace the notification agent with PagerDuty, replace Ollama with a cloud LLM API).

---

🙋 Human-in-the-Loop

PLCpilot is designed with human oversight at every critical step:

• The Human Notification Agent sends alerts before deployment
• Operators can approve or reject generated code
• The Learning Agent flags repeated failures for human review
• All generated code is logged with timestamps for auditability

---

📄 License

MIT License — see LICENSE for details.

---

👤 Author

Suresh Dakha Senior Solution Architect — Physical AI, Edge AI & OT Cybersecurity

Passionate about making industrial automation intelligent, accessible, and secure — bringing GenAI to the factory floor for manufacturers of every size.

---

🔗 More Projects

Explore my other repositories covering OT Security, Edge AI and Physical AI: