Sense → Capture → Insight
We instrument machines that were never designed to be.
Custom sensing hardware, reliable data capture, and models that raise problems before they’re failures. Retrofitted to a lathe built in 1985, or designed into the machine you ship next year.
Illustrative trace. Shaded band is the learned healthy signature; the alert is drift out of it.
The capability chain
Three legs. The first one is the one nobody else will touch.
Make the measurement exist
Off-the-shelf sensors where they fit. Custom capture hardware where they don’t — machinery with no digital interface, awkward physical environments, and signals nobody makes a standard part for.
Bearing temperature °C
Motor current A
Line pressure bar
Acoustic signature to 20 kHz
Get it off the machine intact
Data has to exist before AI can touch it. Reliable acquisition, transport and storage from the shop floor — timestamped at the device and buffered to flash, so a dropped link costs you latency rather than evidence.
Zigbee · meshed across a site
LTE-M / NB-IoT · no site network
CAN / OBD-II · vehicles and plant
Say something useful about it
Baseline the healthy signature, then alert on drift. Bearing wear, imbalance and looseness each have characteristic frequencies; an operator already hears some of it when he says she doesn’t sound right.
Degradation · trend over months
Threshold · with the context around it
Comparison · machine against machine
Where we start
At the machine that can’t produce any data yet. Plenty of firms will chart a machine that already talks. Almost nobody will instrument one built in 1985 with no digital interface and no intention of acquiring one. That leg is the whole company — the sensor, the board, the firmware and the link.
What we build with
Electronics R&D, end to end.
Schematic and board through firmware to the analysis. One outfit, so nothing gets lost between the sensor and the answer.
Current clamp on the supply, accelerometer on the housing, thermal, acoustic. None of it needs the machine’s cooperation or its controls.
ARM, with quiescent current treated as a headline specification rather than an afterthought. Devices that live on a battery for a season, not a weekend.
BLE, Zigbee, Wi-Fi, LTE-M and NB-IoT; CAN and OBD-II. The link gets chosen for the site, not for the datasheet.
Readings timestamped at the device, buffered locally, attributable to a machine and an hour. A measurement that stands up months later is worth more than the sensor that took it.
TLS with fingerprint pinning, rotating tokens, signed messages. Instrumenting a machine should not open a door into the network it sits on.
Where a phone is already present, it carries the cellular link and the position fix — so the hardware pays for neither, and there is no SIM to feed for the life of the device.
Two ways in
Retrofit is the normal case.
“We run machinery, some of it decades old.”
You want to know a machine is about to fail before it does, without ripping anything out or pausing production to find out. Old machines are welcome — they’re the interesting ones. Clamp the sensing on and an uninstrumented machine becomes a data source in an afternoon.
“We build machines and want monitoring in them.”
Sensing designed into the product rather than bolted on afterwards, as a feature you can sell and a service contract you can renew. Real electronics R&D — schematics, boards, firmware — not an integrator gluing subscriptions together.
Start here
Tell us about the machine.
What it is, what it does when it’s misbehaving, and what you currently write down about it. Plain language is fine — working out what to measure is the job, not the prerequisite.
hello@tinystalk.com