In express transit hubs and e-commerce fulfillment centers, sorting efficiency directly determines operating costs and delivery times. However, traditional automated sorting systems face three major hidden pain points:
Vision inspection relies on backend servers – Barcode scanning, damage detection, and label positioning typically require uploading images to an industrial PC or the cloud. Network fluctuations lead to “packet loss” and increased sorting error rates.
Safety hazards remain invisible – Personnel crossing sorting lines without authorization, failure to wear hard hats, unexpected equipment intrusions – such behaviours lack real-time edge monitoring, and post‑event tracing cannot prevent accidents.
IT and OT are “two separate worlds” – The PLC only executes sorting actions; the server only processes data. An extra gateway is needed for protocol conversion, making troubleshooting difficult and slowing down data closed‑loop control.
No on‑site visual “cockpit” – Sorting efficiency, abnormal packages, and equipment status can only be viewed in the control room. Line‑side operators have no real‑time visibility.
The EdgePLC BL237 series industrial edge AI controller integrates real‑time control, AI vision inference, protocol conversion, and local large‑screen display into one device. With low cost, low latency, and high reliability, it empowers every sorting line with “see‑while‑control, decide‑instantly” capability.
1 TOPS NPU – Built around Rockchip RK3562/RK3562J, quad‑core A53 + M0 coprocessor, providing 1 TOPS of AI compute – sufficient to run lightweight YOLOv5/8 object detection models while maintaining a PLC scan cycle ≤1 ms.
Rich interfaces – 2~3 10/100M Ethernet ports, 2x USB2.0, Mini PCIe (for 4G/WiFi expansion), and up to 32 N‑series I/O modules (DI/DO/AI/pulse, etc.). It can interface with photo‑electric sensors, motor drives, pneumatic components, barcode readers, and more.
HDMI 2.0 output – Supports 1080p@120fps or 4K@60fps, directly driving a large‑screen industrial display without an extra industrial PC or set‑top box.
Pre‑installed Ubuntu 20.04 + YOLOv5/8 + OpenCV – Out‑of‑the‑box deployment of models for package detection, label localisation, hard‑hat recognition, etc.
Supports CODESYS / OpenPLC / NexPLC – IEC 61131‑3 compliant, allowing PLC engineers to quickly write sorting logic.
Built‑in BLIoTLink protocol converter – Supports Modbus, OPC UA, MQTT, S7 and other protocols, seamlessly connecting to WMS, MES or cloud platforms.
Node‑RED visual orchestration – Low‑code construction of sorting data flows, e.g. converting AI detection results into PLC control signals.
Pain point – Dirty labels, wrinkled labels, or tilted packages cause high misread rates with traditional fixed barcode readers.
EdgePLC solution
An overhead industrial camera connects to EdgePLC via USB or Ethernet. Each frame is fed into a YOLOv5 model that simultaneously detects the barcode area, package shape, and label orientation.
NPU inference results (decoded barcode information, package centre coordinates) are written directly into PLC variables.
The PLC controls the wheel sorter or cross‑belt to divert the package to the correct chute. If the barcode cannot be recognised, an alarm is triggered and the package is diverted to the reject chute.
Value – Recognition rate increases from ~95% to >99.5%, cloud‑independent, per‑frame processing <50 ms.
Scenarios
Personnel crossing a sorting line safety barrier
Operator not wearing a hard hat
Equipment door left open
EdgePLC solution
Ordinary network cameras on both sides of the sorting line stream live video to the EdgePLC’s NPU.
Detection models for intrusion, hard‑hat compliance, equipment status etc. run on the NPU. Once a violation is detected, millisecond‑level actions are triggered:
Audible/visual alarm (via DO module controlling a beacon)
Sorting line stop (PLC logic immediately stops the conveyor)
Snapshot saved and pushed to the management platform via MQTT
Value – A true “AI security guard” on duty 24/7, reducing accident response time from seconds to milliseconds.
Traditional approach: Barcode reader → industrial PC → gateway → PLC → actuator – at least three protocol conversions.
EdgePLC solution
A single controller runs AI vision (Python/C++), PLC logic (CODESYS), protocol conversion (BLIoTLink) and data upload to cloud (MQTT).
PLC variables can be directly read/written by Node‑RED or Python scripts – no OPC server needed.
Sorting data (each package’s barcode, sorting time, chute number) is written in real time to a local SQLite database and simultaneously pushed to the enterprise ERP or sorting management system via 4G/WiFi.
Value – Hardware cost reduced by ~40%, data latency cut from seconds to milliseconds, and far fewer failure points.
Pain point – Line‑side workers cannot see current sorting efficiency; managers frequently need to enter the control room.
EdgePLC solution
Directly connect a 55‑inch industrial large screen via HDMI 2.0.
Use the built‑in Grafana or FUXA Web SCADA to create real‑time dashboards showing:
Current sorting speed (items/hour)
Accumulated package count per chute
Number of AI‑detected abnormal packages with image carousel
Equipment running status (green / yellow / red)
The large screen also displays safety alerts (e.g. “Intrusion at lane 3 – emergency stop activated”).
Value – Line‑side personnel have immediate visibility, managers can grasp the overall situation during walk‑throughs, improving management efficiency and worker engagement.
| Feature | Traditional solution (PLC + industrial PC + gateway) | EdgePLC BL237 all‑in‑one |
|---|---|---|
| Hardware cost | 3~4 devices, ~5000‑8000 RMB | Single controller, ~2000‑3000 RMB |
| AI vision latency | 100‑300 ms (including network transmission) | <50 ms (local inference) |
| Data protocol conversion | Requires separate gateway or software | Built‑in BLIoTLink, supports 20+ protocols |
| On‑site visualisation | Needs extra industrial PC or set‑top box | Direct HDMI large‑screen connection, no extra device |
| Maintenance complexity | Multi‑vendor integration, difficult fault isolation | Single brand, web‑based remote O&M (BLRAT) |
| Environmental robustness | Fan‑cooled industrial PC prone to dust clogging | Fanless, -20℃ to +85℃ wide temperature, IP30 |
[Industrial camera] --USB--> EdgePLC BL237 (YOLOv5 barcode + package detection) | |-- HDMI --> On‑site 55" large screen (live dashboard) | |-- DO module --> Alarm beacon / emergency stop | |-- EtherCAT --> Cross‑belt servo drive | +-- 4G module --> Enterprise cloud platform (sorting reports)
Program logic (CODESYS + Python hybrid):
Python thread: read camera frames → NPU inference → obtain barcode & coordinates → write to shared memory.
PLC cycle: read from shared memory → calculate sorting chute based on coordinates → send pulse commands to cross‑belt → count and upload.
The EdgePLC BL237 industrial edge AI controller integrates AI vision inspection, safety hazard detection, real‑time PLC control, IT/OT data fusion, and on‑site HDMI large‑screen display – replacing three devices with one. It offers a low‑barrier, high‑real‑time, strong‑intelligence new approach for logistics sorting automation.
It stops sorting lines from “running blind”, leaves unsafe acts “no place to hide”, and gives managers “at‑a‑glance” insight – exactly what the next generation of smart logistics sorting systems should look like.
