Our AI sorting system identifies, grades and diverts goods as they move — backed by edge-AI controllers, machine vision and distributed I/O built for production floors, where a 200 ms round trip to a data centre is not an option and a stopped line costs real money.
Cloud inference adds latency you cannot schedule around. Industrial PCs are not built for 24/7 vibration and 55 °C cabinets. And integration usually means three vendors pointing at each other when the line stops.
Our automation products put the model, the motion control and the fieldbus on one board we designed — so the decision is made in the same cycle as the movement, and one team is accountable for the result.
Deploy one, or the whole set. Each is available as a standard product or engineered to your specification.
The flagship of this line. Goods arrive on the conveyor unsorted; the system identifies each one by type, grade, size, colour or defect and diverts it to the right lane — at line speed, with a logged record per unit. Trained on your own products rather than a generic dataset, and retrainable on-site when your mix changes.
Handles mixed-SKU streams, damaged-goods rejection, batch separation and count verification, and reports what it rejected and why — so a drift in your upstream process shows up as a number instead of a customer complaint.
Identify · grade · divert · logFanless DIN-rail controller running your inference model alongside a real-time control task. Multi-core SoC with NPU, isolated digital I/O, dual GbE and CAN — the brain of a cell or a whole line.
Model + motion + fieldbus in one boxInspection stations for surface defects, presence/absence, OCR and dimensional checks. Trained on your own defect library, retrainable on-site as your process drifts — no cloud upload of production images.
AOI · OCR · MetrologyEtherCAT and Modbus remote I/O blocks, stepper/servo drive interfaces and safety-rated inputs — so sensors and actuators at the far end of the line stay in sync with the controller.
EtherCAT · Modbus · Safe I/ONavigation and fleet-control hardware for warehouse robots — LiDAR SLAM, obstacle avoidance, docking and charging management, with a fleet API for your WMS.
SLAM · Fleet API · DockingIndicative figures for the standard variant. Every parameter is configurable on an OEM build.
| Parameter | Specification |
|---|---|
| Compute | Multi-core Arm SoC with integrated NPU; optional FPGA co-processor for deterministic I/O and custom pipelines |
| AI runtime | INT8 / FP16 quantised models; ONNX and TFLite import; on-device retraining pipeline available |
| Vision input | Up to 4× GigE Vision or MIPI CSI-2 cameras; hardware image signal processing |
| Fieldbus | EtherCAT, PROFINET, Modbus TCP/RTU, CANopen; OPC UA and MQTT northbound |
| I/O | Isolated digital in/out, analog in, encoder inputs, RS-485/RS-232; expandable via remote I/O |
| Operating system | Embedded Linux with real-time patch; containerised application deployment |
| Environment | −20 °C to +60 °C operating, fanless, vibration-rated, DIN-rail or panel mount |
| Power | 12–36 VDC industrial input with reverse-polarity and surge protection |
| Certification | Designed for CE / FCC EMC; additional market certification to project scope |
Specifications are indicative and subject to change. Contact us for the current datasheet of a specific model.
Products separated by type, grade and destination on a shared line, with a per-unit record for the WMS.
Solder joint and connector inspection at line speed, catching defects before they reach a downstream station.
Label presence, print quality, fill level and seal integrity checks with reject actuation on the same controller.
Dimensional verification and surface defect grading, with traceability records pushed to MES over OPC UA.
AMR navigation, docking and charging control, plus sortation and barcode reading at the induction point.
Vibration and current signature analysis on the edge, alerting before a bearing or spindle takes the line down.
OEE, cycle-time and downtime capture aggregated at the cell and published northbound to SCADA dashboards.
We study the process, cycle time, defect classes and existing PLC landscape.
Sample images and data are used to prove detection rates before hardware is committed.
One station deployed and tuned against your real production mix.
Replication across stations or plants with a repeatable configuration.
Model retraining, spares and lifecycle support through the equipment's life.
Describe the process, the cycle time and what needs to be detected or controlled. We will propose the hardware, the detection approach and a pilot plan.