Natural Language Industrial Control

Intelligent Industrial Automation Platform: Integration of Edge AI, IoT, and Cloud Supervision
Abstract
This project represents a modern approach to industrial automation, bridging the gap between traditional PLC (Programmable Logic Controller) systems and advanced AI/IoT technologies. It consists of a distributed architecture featuring an intelligent edge node (Raspberry Pi), a central coordination server, and a high-performance React-based HMI (Human-Machine Interface). The system allows for real-time control, data logging, and novel 'Natural Language Control' capabilities, enabling users to interact with industrial hardware using conversational AI

System Architecture
The project is divided into three main components:

1. Intelligent Edge Controller (ServidorPlc)
Hosted on a Raspberry Pi, this component acts as the "Soft PLC".

Technology: Node.js, Express, Socket.io, Node-OPCUA.
Functionality:
Hardware Interface: Direct actuation (PWM, Digital I/O) and sensing (ADC communication).
Control Loops: Implements PI/PID algorithms and automatic system identification (ARX models).
AI Integration: Processes natural language prompts (via OpenAI/Local LLMs) to generate control commands dynamically.
Connectivity: MQTT and WebSockets for real-time telemetry.
2. Supervisory Interface / HMI (Plc)
A web-based SCADA/HMI application for monitoring and control.

Technology: React, Vite, CoreUI, Redux, Vega-Lite.
Features:
Data Visualization: Real-time charts (Vega-Lite) for sensor data (temperature, level, flow).
Command Center: Interface to send voice/text commands to the PLC.
Dashboard: CoreUI-based administrative panels for user management and configuration.
3. Backend & Data Layer (ServidorPc)
(Contextual) A central server or database handler for long-term storage and aggregation.

Technology: Node.js, PostgreSQL.
Functionality:
Data persistence for historical analysis.
Centralized authentication and user management.
Key Features
Automatic Model Identification: The system can automatically perturb the plant, record response data, and fit ARX (AutoRegressive with Exogenous input) models (Order 1-3) to mathematically represent the system dynamics.
Generative AI Control: Users can issue commands like "Maintain level at 50% indefinitely" or "Identification test with steps of 5 seconds," which are parsed into executable machine code.
Remote Telemetry: Real-time streaming of process variables (Setpoints, Process Values, Control Outputs).
Extensibility: Modular Node.js architecture allows easy addition of new protocols (Modbus, OPC UA) or AI models.