Terrafa.
Request access Contact us
/News /Terrafa engineering
TF-HW-001 · Bench test record · Hardware programme

First bench test of an early AI-embedded hardware prototype, with readings streamed live into Continuum Core

James Parkinson1, Orsell Biokou2 & Beatrice Hammond3

1Technical Engineering Lead and Co-founder, Terrafa Limited. 2Co-founder and Chief Executive, Terrafa Limited. 3Software Engineering Lead, Terrafa Limited.

Test reference
TF-HW-001
Date
29 August 2026
Duration
4 hours
Status
Complete
Data retained
Yes, held in Continuum Core

Hardware that reads a product where it is stored lets a business see its stock at every site, and it saves the time staff spend collecting readings by hand. Terrafa is developing its own AI-embedded hardware, built on infrared spectroscopy and graphene, and published work shows that infrared spectroscopy can determine the chemical composition and physical properties of diesel1. This record describes Terrafa's first laboratory test of an early prototype, which read a diesel sample and sent every reading wirelessly to Continuum Core in under a second. The aim was to show that readings from a diesel sample can arrive in Continuum Core continuously for a set period. The test ran for one hour. Further tests will find how long the hardware runs before a fault.

Test set-up

A diesel sample was collected at the pump and stored in an airtight aluminium container. The prototype read the sample on the bench once a second. Each reading travelled as a single message over a private network connection to Terrafa's data store, and from there into Continuum Core.

Fig. 1Data path. Readings travel from the hardware over a private wireless network into the Data Feeds service, and arrive as timestamped rows in Continuum Core.
Drawing of the bench: a capacitive probe standing in a beaker of sample at left, a beaker seated in a timber block between an emitter and a detector in the centre, and an ESP32 board with a small display at right, wired to both.
Fig. 2Bench set-up. The early prototype on the laboratory bench during the test.
Drawing of the optical cell: a beaker seated in a machined recess in a timber block, with the IR emitter on the left face and the IR detector on the right face so the beam crosses the sample.
Fig. 3Prototype detail. A closer view of the prototype on the bench.
ItemDescriptionAmountNote
SampleDiesel collected at the pumpOne sampleKept in an airtight container
HardwareEarly AI-embedded hardware prototypeOne unitBench test only
ConnectionPrivate wireless network to Continuum CoreLiveNo manual transfer

Table 1Test summary. The sample, the hardware and the connection used in this session.

Results

Local display

The local readout during the run is shown in Fig. 4.

Photograph of the small OLED display on the bench, lit with three lines of readings, among the harness wiring.
Fig. 4Local readout during the run. The display updated with each new reading throughout the run.

Readings

The run is summarised in Table 2. Every reading arrived in Continuum Core.

Run itemUnitStartEndNotes
Run timeminutes060One continuous run
Reading rateper second11
Messages lostcount00
DeliverymodeLiveLive

Table 2Run log. This table gives the run time, reading rate and delivery for this session.

Arrival in Continuum Core

Rows arrived in the Data Sources catalogue while the hardware ran (Fig. 5).

Photograph of a laptop running Continuum Core on its Data Sources tab, with a live feed listed on the right and a table of timestamped readings filling on the left.
Fig. 5Continuum Core, Data Sources tab. The live feed appears as a dataset with its own header, and readings fill the table as timestamped rows.

Observations

The prototype ran for one hour, and no messages were lost. Rows arrived in Continuum Core one by one as each reading was taken, so the connection carries a live feed.

Development

The prototype sends its readings over a private wireless network to Continuum Core. The data path from sample to database is shown in Fig. 1.

Hardware

The team built a bench fixture that holds a sample in the same place for every run. It is the first fixture Terrafa holds for repeatable bench tests, and the prototype shows its readings on a local display.

Firmware

The firmware takes each reading, shows it on the local display and sends the same value to Continuum Core. It is written in C++.

Software

The Terrafa Data Feeds application receives readings from the hardware and streams them to the Continuum client. Both applications are written in C#, with samples available in the Continuum Data Feeds repository on GitHub.

Conclusions

  • An early AI-embedded hardware prototype can run and send readings continuously on the bench.
  • Readings pass from the hardware into Continuum Core as a live dataset, and nobody has to move data by hand.

Future work

  • Controller. The bench test used a general-purpose controller. Longer tests will run on industrial-grade controllers built for continuous use.
  • Terrafa's own hardware. This early test used standard parts. Terrafa is developing its own AI-embedded hardware, built on infrared spectroscopy and graphene, for use at customer sites.

Photograph log

Figures 2 to 5 form the photograph log for this session.

References

  1. Hradecká, I., Velvarská, R., Dlasková Jaklová, K. & Vráblik, A. Rapid determination of diesel fuel properties by near-infrared spectroscopy. Infrared Phys. Technol. 119, 103933 (2021). https://doi.org/10.1016/j.infrared.2021.103933
All news Continuum Sense