● Telemetry · Industry 4.0

Machine data that stays on the plant LAN.Sub-50 ms loops. Seventy-two hours if the WAN drops.

IIoT Edge AI filters spinning, knitting, and weaving signals on the gateway. It speaks OPC UA and MQTT, buffers up to 72 hours, and can still act on the line when the internet is down.

eManage textile IIoT edge processes machine telemetry locally in 10–50ms, speaks OPC UA and MQTT, retrofits Modbus/RS-485 frames, and buffers 24–72 hours so OEE and stop relays survive internet outages before syncing to ERP.

Loop

<50 ms

Stop & tension relays

Buffer

72 h

Circular, zero loss

Legacy

Modbus / 485

Frames without Ethernet

WAN

−99.8%

Edge rollups, not raw

Plant-LAN edge gateway — Jetson-class inference beside the machine cabinet
TIER 1–4 · OPC UA · MQTT

Why edge

Cloud ingestion is too slow and too expensive for a shed.

Hundreds of spindles at 100ms, looms above 1,000 picks/min. A 200–500ms cloud hop cannot hold a stop relay. An outage blanks the dashboard.

200–500ms

Cloud RTT

Too sluggish for tension and halt decisions on the frame.

120k/min

Raw points

200 spindles × 10 readings/s before anyone aggregates.

WAN down

Blind mill

Cloud-only stacks go dark the moment the fibre link drops.

On the box

Four monitoring lines. One edge write-back.

Same gateway, same 72-hour buffer. Knit cells, looms, sewing lines, and machining centres each have their own console on the same edge path.

Circular knitting machine with yarn creels — RPM and yarn-break telemetry on the cell gateway

Circular & flat knit

Knitting Monitoring AI

Pick rate, yarn-break, and stop reasons leave the circular-knit or flat-knit cell in 10–50 ms. The edge box writes the live job card — not a clipboard at shift end.

  • Yarn-break and needle-line stops
  • Pick / course rate vs the ticket
  • Store-and-forward if WAN dies
Open the knit-cell console →
Weaving loom with warp sheet and fabric — loom counter and stop telemetry on the shed gateway

Looms & weaving sheds

Weaving Monitoring AI

Loom counters and stop signals leave every loom in 10–50 ms. Opening and closing readings, net metres, and weaver averages fill the Day and Night shift report — no walk round the shed with a pen.

  • Net metres and kg per loom, per shift
  • Warp, weft, beam, and fitter stops with minutes
  • Production, finishing, and fabric stock in the ERP
Open the loom-shed console →
Operator at a lockstitch machine on a garment sewing line — SPM and needle time on the line gateway

Garment sewing lines

Sewing Monitoring AI

Servo-motor taps and clamp-on sensors stream SPM, needle time, and idle reasons from every station. The hour-by-hour board and the bundle card update while the style is still on the line.

  • SPM and needle time per station
  • Bottleneck and WIP flagged live
  • Piece-rate and DHU on the bundle card
Open the sewing-line console →

Protocol matrix

What runs at each hop

Bandwidth, latency, and encryption change from the motor to the HO ledger. The gateway is the translator, not a cloud pipe.

MQTT

Edge to plant / cloud

Publish/subscribe with TLS 1.3. Spindle speeds, picks, and power with QoS 0–2. Typical loop 10–50 ms.

OPC UA

Machine to gateway

IEC 62541 on Rieter, Picanol, Murata, Toyota. Information models, X.509, AES-256. Sub-10 ms polling.

Modbus

Legacy frames & meters

TCP/RTU on older spinning, carding, and energy meters. Isolated VLAN — never exposed to the office LAN.

HTTPS

ERP & BI sync

Hourly tallies, tickets, and inventory into eManage. Batch, not a tension loop.

How it works

01

Sense

Current transducers, pick pulses, inverter frequencies — including RS-485 breakouts on machines that never had a port.

02

Filter on the box

Local averages, alarms, and OEE. 90–98% of raw volume never leaves the shed.

03

Act

Relays and operator lights fire from the gateway, not from a region.

04

Forward

When WAN returns, alarms and KPI rollups go first; raw streams follow. ERP catches up in priority order.

Four-tier edge

Where the millisecond lives

T1

Machine

Pick pulse, inverter, RS-485 on the frame.

T2

Cell gateway

OPC UA / MQTT, <50ms loop, 72h buffer.

T3

Plant

Shed HMI, line rollups, local alarms.

T4

ERP

eManage sync — KPIs first, raw later.

The same numbers payroll uses

Shift efficiency and stop reasons land on job cards. Piece-rate and OEE stop arguing because they are no longer two notebooks.

  • Rieter, LMW, Schlafhorst, Picanol, Sulzer — mixed in one model
  • AES-256 at rest, mTLS 1.3 with device certs in flight
  • VLAN isolation, locked USB, locked enclosure

Four tiers

  • Tier 1 — machine

    Sensors and serial on the frame. Spindle FFT, loom pick, warp loadcell, IO-Link / 4–20 mA / RS-485.

  • Tier 2 — cell gateway

    Ruggedized OPC UA / MQTT box, 72-hour circular buffer, local departmental broker.

  • Tier 3 — plant

    On-prem GPU/NPU: yarn-break prediction, shed OEE, local HMI.

  • Tier 4 — ERP / HO

    eManage sync, multi-plant rollup, long-term telemetry lake.

Questions plants actually ask

Latency and cost. Relays need 10–50ms. Raw spindle streams also burn ingress. Edge filters to machine averages and department OEE first.

Local alarms and models keep running. A 24–72h circular buffer holds telemetry. On reconnect, alarms sync first, then raw history.

Typically 90–98%. A 200-spindle example drops ~120,000 points/min to a few hundred machine averages and OEE metrics.

RS-485, Modbus RTU/TCP, and analog breakouts convert to OPC UA/MQTT so old and new machines share the same dashboards.

Physical locks, industrial VLANs, disabled USB, AES-256 at rest, mutual TLS 1.3 with per-device certificates.

Next