Predict at the edge,
decide on the screen
An industrial edge HMI that runs a lightweight ONNX inference engine on a Raspberry Pi 4/5 industrial touch monitor to perform operation control, real-time monitoring, and AI predictive maintenance for temperature/humidity chambers, HVAC, and precision environmental chambers on a single device. Decide instantly on the floor without cloud dependence, and deploy without downtime via OTA.
On a Raspberry Pi-based 800×480 industrial touch monitor, temperature/humidity, equipment health, and operation control are shown on a single screen. This is an actual operations screen.
When deployment is typically considered
The environment makes cloud connectivity difficult
An isolated factory network and poor line conditions make relying on an external cloud difficult, yet equipment anomalies must be judged instantly on the floor.
Alarms come in, but the cause is unknown
Even when high- or low-pressure alarms trigger, only context-free numbers appear, so staff must check in person whether it is a compressor fault or a refrigerant shortage.
Replacing the existing control panel is a burden
A retrofit approach is needed that leaves the already running microcontroller and control logic untouched and only adds an AI diagnostics layer on top.
Five core capabilities
From AI inference, touch screens, and field communication to unattended deployment and edge autonomy, every stage of an industrial edge AI HMI is delivered on a single device.
Edge AI inference — decide instantly on the floor
A 117K-parameter lightweight ONNX model runs directly on the device. Real-time inference is possible even on a 2GB RAM Raspberry Pi 4, so diagnostics and alerts do not stop even if the network drops. A 256-point sliding window and EWMA smoothing produce a noise-robust anomaly score, and when warning or danger thresholds are exceeded, it alerts along with the primary cause.
- onnxruntime · FP32 ONNX
- 256pt sliding window
- Warning 0.45 / Danger 0.75 thresholds
- Primary cause display
Touch HMI — an operator-centric full-screen interface
A PySide6-based kiosk UI optimized for an 800×480 industrial touch screen. It delivers the dashboard, operation status, graph, alarm history, AI diagnostics, settings, and device info screens as a single app, designed for one-finger operation.
- PySide6 / pyqtgraph
- Kiosk full-screen
- X11/Wayland compatible
Field communication & control — connect instantly to existing equipment
It supports direct RS-485 Modbus RTU connection (MICOM microcontroller), a FastAPI-based internal Bridge API, and an MQTT 5 cloud bridge as standard. Retrofit deployment is possible by adding only an AI diagnostics layer without replacing the existing control panel or microcontroller.
- RS-485 Modbus RTU (FC03/06)
- FastAPI Bridge API
- MQTT 5 cloud bridge
- Retrofit
Hybrid OTA — safe, unattended deployment
When firmware is pinned per device and per model on the server, the agent downloads it automatically and then verifies the Ed25519 signature and SHA-256 integrity. When the operator approves [Update Now] on the screen, automatic installation and a health check proceed, with automatic rollback on failure. Unintended deployment is blocked at the source, and updates can be applied safely even during LIVE operation.
- Ed25519 signing · SHA-256 verification
- Operator approval gate
- Health-based automatic rollback
Edge autonomy — normal even when the cloud is disconnected
Even if the internet drops, inference, control, logging, and alerts continue to work normally, and once restored, the accumulated decision log is automatically synced with the cloud. It is robust against isolated factory networks and poor line conditions.
- Offline autonomous operation
- Automatic decision log sync
- Isolated-network support
Screen tour
These are actual operations screens delivered by the single EGNOX AI Mon app. They are based on a temperature/humidity chamber AI fault-prediction configuration, and the displayed items are configured to match the deployed equipment.
* These are actual operations screens; the displayed items and configuration may vary depending on the deployed equipment and requirements.
Tech stack
| Hardware | Raspberry Pi 4 (2GB, Py 3.9) / Raspberry Pi 5 (8GB, Py 3.11) + 800×480 industrial touch (CPi-S070WR, etc.) |
|---|---|
| AI runtime | onnxruntime + FP32 ONNX (optimized for ARM Cortex-A72 / A76) |
| HMI | PySide6 / pyqtgraph (X11/Wayland compatible) |
| Communication | RS-485 Modbus RTU (FC03/06), MQTT 5, HTTPS (OTA) |
|---|---|
| OTA | FastAPI server + Ed25519 signing + health-based automatic rollback + hybrid approval |
| Operations | systemd auto-start on boot + journal unified logging + decision log upload |
Applications
Temperature/humidity chambers · Precision environmental chambers
Detect compressor faults, refrigerant shortages, and high/low pressure early, and operate temperature/humidity settings and PID control from a single screen.
HVAC · Air-conditioning equipment
Diagnose compressors, humidifiers, and blowers in one place, and check operating status and early signs instantly on the field touch screen.
Small edge-equipment retrofit
Add AI diagnostics to existing microcontroller equipment without modifying the control logic. Connect directly via RS-485 and deploy without replacing the control panel.
Operating many distributed units
Manage tens to hundreds of units unattended in bulk via OTA. Update safely with per-device and per-model pinning and health-based rollback.
AI, screens, communication, and OTA
in one solution
It integrates every stage of an industrial edge AI HMI into a single device, simplifying deployment and operation.
When a new AI model or feature is released, it is applied to the floor directly via OTA.
Data learned on the floor → cloud retraining → automatic application via OTA.
Built on a low-cost SBC, it consolidates the PLC, HMI, gateway, and AI box into one.
Products worth considering together
Tell us about your equipment
and communication environment first
After confirming the equipment type (temperature/humidity chamber, HVAC, chamber, etc.) and the microcontroller communication method, we will organize the retrofit configuration and the screen and OTA operation approach together.