Skip to content
IEMI Labs
All product designs
IndustrialDesign rev A, 27 Oct 2026: CAD, fit checks, assembled moulding-cell model and 5 shared checks done; not built

LineCell

One cell kit for a moulding machine: checks every part and watches the machine that made it

LineCell does the jobs of QualityCheck AI (teach-by-example PASS / FAIL / UNSURE on every part with a reject output and lot log) and PlantNode (machine running or stopped from the stack light, pump-motor bearing trend, air in dollars, hot cabinet joints and the machine's own dials) at one machine cell, with one brain: the IEMI Edge Hub at the inspection bench. The PlantNode head on the machine cabinet now also does the QualityCheck I/O pod's job and talks to the hub over one USB lead. Because one brain sees both, every part verdict carries the machine state, so scrap after a restart and reject spikes that line up with hot oil or low air are named automatically.

Who buys it
The owner, production manager or QC lead at a 10-200 person injection moulder or small parts maker, who today inspects by eye and finds out about machine trouble from the scrap bin.
Why it pays
Bought separately, QualityCheck Inspect ($3,499 + $399/mo) and PlantNode ($1,190 + $39/mo, plus a $399 LoRa gateway) cost $5,088 + $438/mo for one cell, with two apps that never meet. LineCell is $3,990 + $349/mo (target) with one hub, one app and no gateway.
Target price
$3,990 per cell + $349/month (target), after the $500 feasibility study on your parts (credited)
First-minute wow
The app says '9 parts failed after the 10:40 stop, and the spike at 14:05 came while the oil dial read 58 °C', not just '18 fails today'.
LineCell concept: the idea and where each part goes
The concept. The CAD design is below, under Printable CAD.

What the research found

And how this design answers it

  • Two products, two apps → one cell kit with one Edge Hub brain that runs the part check and all five PlantNode jobs
  • The QualityCheck I/O pod and the PlantNode head merged: one ESP32-S3 head on the machine cabinet with 24 V trigger and reject terminals and the machine sensors, one USB lead to the hub
  • Dropped PlantNode's LoRa radio, $399 site gateway and 5 V panel supply; one 24 V adapter powers the head, LED ring and outputs
  • New shared jobs (5/5 checks pass): every verdict tagged with machine state; restart scrap counted; reject spikes tagged with oil temperature or air pressure; energy per good part from the pump clamp; a weekend leak test of the cell's own air using a reject-style output on a drop valve
  • Rules: monitor and alert only; LineCell never stops the machine, and work inside the cabinet is for a qualified electrician

Use case

A 120 t moulding machine making housings, with a QC bench beside it

Production managerSetterQC inspector

  1. 01Trigger: each housing drops down the chute; the inspector places it in the nest and the photo-eye fires
  2. 02Detection: the hub scores the part in about 0.4 s and joins it to the machine's state, shot count, oil and air readings
  3. 03Alert: after a 10:40 stop the first 9 parts fail and are logged as restart scrap; at 14:05 a spike shows with the oil dial at 58 °C
  4. 04Action: the setter checks the oil cooler before more scrap is made; on Monday the app shows pump bearings and the cell's leak cost

Outcome: Bad parts caught at the bench and their cause named from the machine, from one kit and one app

How it works

Architecture

  1. 01

    Sense: light-box camera; stack-light clip, pump puck, current clamp, air-drop pressure, cabinet thermal eye and joint tags, dial camera

  2. 02

    Decide: QualityCheck part check on the Hailo-8L; PlantNode jobs and the shared cell report on the Pi 5

  3. 03

    Send: the head talks to the hub over USB; the hub sends to the app over Ethernet and MQTT

  4. 04

    Tell: operator screen at the bench, one LineCell app, SMS for stops, hot joints and reject spikes

Hardware

Real parts, buyable today

Bill of materials: about $1,420 per cell at qty 1 (est.).

BlockPartWhy this part
Edge Hub (the one brain)Raspberry Pi 5 8 GB + AI HAT+ 13 TOPS (Hailo-8L)Only part that can run the vision model and every machine job
Light boxGlobal Shutter camera, 16 mm lens, strobed LED ring, part nest, photo-eyeUnchanged from QualityCheck
Cell headESP32-S3-WROOM-1-N16R8, OV5640 dial camera, ADS131M04, M7E-Hecto RFID, 24 V I/O terminalsPlantNode head plus the QualityCheck I/O pod
Machine sensorsIIS3DWB puck, SCT-1250 clamp, PX3 pressure, MLX90640 eye, AZN3254 tags, stack-light clipUnchanged from PlantNode

Mechanical

Printable CAD

14 parts, generated from code and checked as closed solids. Pick a part to inspect it, or download the STL files to print.

Loading the CAD…

Real CAD · drag to rotate · scroll to zoomReal CAD · swipe sideways

Open risks we are working on

  • Restart-scrap and spike causes are correlations, not proof; the setter decides what to change
  • USB serial over 3 m next to a moulding machine's drives: noise test needed (fallback RS-485, est. +$6)
  • Not every moulder has an analog oil gauge or a separate air drop; the dial and leak jobs depend on the machine
  • Big press makers sell their own connected controllers (e.g. Engel e-connect, Arburg ALS) and big vision makers sell inline cameras; our buyer is the small, mixed-fleet shop
  • Price $3,990 + $349/mo against a BOM of about $1,420 (est.) leaves room, but the feasibility study must convert

Design stage only: nothing on this page is for sale. Specifications and prices are targets.