01
Configure the AI
Point it at the machine. The autoencoder learns that asset's own normal — no expert spectra setup.
Runs on industrial PCs
Own Your Edge.
On Beckhoff hardware you own. No Subscription. No Cloud Lock-In. From threshold alarms to edge AI, on Data with real meaning.
- hardware availability
- 20+ yr
- data ownership
- 100%
- acquisition cost
- One-off
hardware availability
data ownership
acquisition cost
Wired · EtherCAT
Know a bearing is failing three weeks before it does
Autonomous edge AI scores a 0–100 Machine Stress Index from 24-bit, 50 kS/s raw capture — alarms and physical I/O react in sub-seconds, no cloud round-trip.
Northbound on open protocols
For plant managers
The Obsolescence-Proof Edge
Sold, versus needed.
Legacy systems
- ~5-year obsolescence cycles
- Proprietary firmware lock-in
- Manual expert configuration per asset
- Vendor-controlled data silos
SensorDataFabrics
- 20+ year hardware availability
- Open PC-based architecture
- Autonomous edge AI that learns the machine
- 100% data ownership, open protocols
Let the AI find it. Then prove it yourself with the raw data.
The autoencoder watches continuously and flags the anomaly; the post-processing suite is how you diagnose it — on months of full-rate captures.
Long-horizon logging on the node — compression-first storage keeps months of history, so the evidence is still there when the AI points at it.
02
The AI flags it
Machine Stress Index climbs out of band. Watch → Warning → Danger, locally, in under a second.
03
Then you investigate
Open the captures around the event and run the post-processing that answers why.
Post-processing toolkit — on demand, on the recorded data
FFT spectrum
harmonics, BPFO / BPFI lines
STFT waterfall
how the spectrum evolved
Power spectral density
broadband energy
Demodulation
envelope, early bearing defects
Cepstrum
sideband families, gear mesh
Spectral kurtosis
where the impulsiveness lives
Time domain
the raw waveform itself
Official ecosystem hardwareBacked by Beckhoff, not by us alone
Every SensorDataFabrics node runs on a standard Beckhoff industrial PC — CX9240 to C6030 — with TwinCAT and EtherCAT underneath. If we disappeared tomorrow, your hardware, spares and support contracts would still come from a global automation vendor you can already buy from.
- 20+ yr
- documented long-term IPC availability
- €1.75 bn
- annual revenue, 5,700+ staff worldwide
- 75+
- countries with local Beckhoff support
- 1980
- PC-based control pioneer, still independent
TwinCAT 3 runtimeEtherCAT · 65,535 nodesDistributed Clocks · sub-µs syncFSoE SIL 3 safety optionOff-the-shelf lead times
Open by design
Nothing proprietary between the sensor and your SCADA
Built on standards you already run.
Two ways to wire the plant. One architecture.
Pick wired or wireless per site — how many sensors you can run comes down to the Beckhoff IPC you put behind them.
Wired
TwinCAT · EtherCATDaisy-chain the edge. Own every revolution.
- 24-bit raw capture at 50 kS/s, continuous
- Deep DSP: FFT, waterfall, cepstrum, demodulation
- Hard-wired I/O and sub-second machine protection
Sensor capacity by IPC
- CX9240
- ~12 sensors · ~6 with DSP + AI
- C6015 / C6025
- ~40 sensors · ~20 with DSP + AI
- C6030
- ~120 sensors · ~60 with DSP + AI
Wireless
LoRaWAN · ChirpStackCut the conduit. Own the airwaves.
- Kilometres of coverage from a single gateway
- Raw waveform and DSP on a duty-cycled interval
- No cable pulls on remote or rotating assets
Sensor capacity by IPC
- CX9240
- ~24 sensors · ~12 with DSP + AI
- C6015 / C6025
- ~80 sensors · ~40 with DSP + AI
- C6030
- ~240 sensors · ~120 with DSP + AI
Six capabilities. Run only the ones the machine earns.
- 01
Sensors
IEPE · wireless
- 02
Beckhoff IPC
CX9240 → C6030
- 03
Edge services
Alarms · DSP · AI
- 04
Your stack
OPC UA · MQTT · Modbus
Core
Metrics & Alarms
RMS · Peak · Kurtosis → tower lights
Optional
High-Speed DSP
FFT · envelope · spectrogram · RPM
Optional
Edge AI
0–100 Stress Index, scored locally
Core
Storage & Egress
Local DB → OPC UA · MQTT · Modbus
Core
HMI & Ops
Web HMI · RBAC · offline rollback
Optional
Fleet Dashboards
Every site, one Azure view
Named, specific, checkable
| Vendor | Their constraint | SensorDataFabrics |
|---|---|---|
| Emerson AMS | 12-channel CHARM limit | 64 sensors per EtherCAT ring |
| Allen-Bradley | 87-day hardware lead times | Off-the-shelf Beckhoff availability |
| Bently Nevada | Manual expert configuration | Autonomous edge learning |
| Siemens | Cloud-only, high-latency models | Sub-second local reaction |
| Augury / Tractian | Proprietary cloud fees, data silos | 100% data ownership, open protocols |
Wired or wireless — same platform
| Capability | Wired · TwinCAT/EtherCAT | Wireless · LoRaWAN |
|---|---|---|
| Real-time metrics & alarms | ||
| Edge AI stress index | ||
| Raw waveform capture | Continuous, high rate | Duty-cycled, fewer points |
| Deep DSP (FFT, envelope, spectrogram) | Yes, on VibraSens sensors | |
| Physical I/O control | ||
| Cable-free retrofit |
Buy it once. A subscription never stops.
One-off hardware and licence. You own it; it runs without us.
- recurring fees, edge
- 0
- optional cloud
- Per sensor
- before it costs a shift
- Catch the failure
recurring fees, edge
optional cloud
before it costs a shift
- SensorDataFabrics — one-off hardware purchase. Optional cloud is the only recurring line, and only if you opt in.
- Typical monitoring subscription — cheap on day one, never stops.
The subscription passes the one-off purchase in year 2 and keeps climbing. Everything after that — plus the downtime the edge AI catches — is margin you keep. The shape is illustrative; your configuration produces a real engineering RFQ.
Prove it on one machine, at our cost
Send your machine list — you get hardware sizing and a payback estimate back.
Or start with a free single-machine pilot on entry hardware — prove it before you ask for budget.
Questions engineers actually ask us
Does this work with our existing EtherCAT masters?
Yes. SensorDataFabrics runs as a TwinCAT application on Beckhoff IPCs and joins your existing EtherCAT segment; measurement terminals daisy-chain onto the same ring, up to 65,535 nodes with sub-microsecond Distributed Clock synchronisation.
What happens to logging during a plant power outage?
Every edge node keeps its own local database. Logging, alarms and AI scoring continue without the network or the cloud, and buffered data is forwarded northbound once the link returns.
What is the real sample-rate ceiling over LoRaWAN?
Wireless VibraSens sensors do capture raw waveforms and run deep DSP — FFT, envelope and spectrogram — but on a duty-cycled interval with fewer data points per capture than the wired path. For continuous raw streaming and sub-second physical I/O reaction, use the wired EtherCAT variant.
Is this a subscription?
No. The edge system is a one-off hardware and licence purchase that you own outright and that keeps running with no connection to us. The optional cloud fleet dashboard is the only item that recurs, and only if you opt in.
How slow can a machine turn before you stop seeing anything?
24-bit resolution lets SensorDataFabrics resolve bearing defect energy down to roughly 12 RPM, where 16-bit systems only see their own noise floor.