Wien, 05.11.2026 (PresseBox) - Nine documented applications. One reusable software core instead of a new AI model for every task.

Calcium. Magnesium. Iron. Blood glucose.

Temperature. Pressure. Rotational speed. Vibration.

The first set could come from a medical lab report. The second from a Formula 1 car, a water pump or a power plant.

Two worlds that appear to have nothing in common. Different instruments, different experts, different physics.

And yet, behind both lies the exact same question:

What state is a system in ? and how is that state changing?

For a piece of software developed in Vienna, both sets of readings present the same task.

A Blood Test Is a Snapshot. But What Happens Next?

It captures a moment in time through many laboratory values. None of those values, taken alone, tells the whole story. A machine is no different. Only when pressure, temperature, rotational speed and vibration are aligned to the same point in time ? or to a defined measurement window ? do they form a coherent snapshot of its operating state.

Semantic Folding Sensor Fusion, or SF2, translates that snapshot into a compact digital State Fingerprint: a sparse binary pattern that can be mathematically compared with other states. The same data, the same model and the same preprocessing produce the same result. Deterministic and reproducible.

And this is where it gets interesting.

One snapshot becomes many. Still frames become a film. Only the sequence reveals how a system is changing.

A single snapshot shows where a system stands.

A series of snapshots reveals how it is changing.

Comparison with known states shows which which patterns it is approaching and which it is drifting away from.

A fingerprint of normal operation makes deviations visible ? potentially before any individual sensor breaches an alarm threshold. If patterns of documented faults are available, SF2 can also measure whether the current state is becoming increasingly similar to them.

This is not fortune-telling. It is a measurable convergence towards known patterns.

?What kind of data we process doesn?t matter to our software,? explains Francisco Webber, founder, CEO and Chief Scientist of SF2 Systems. ?Calcium or crankshaft ? the principle is always the same. We turn many individual values into one state ? and many states into a direction.?

Four Days Before Failure. And Then, Suddenly, Radar.

An industrial pump: 52 sensor channels, more than 220,000 data points and seven documented fault phases. In a retrospective analysis, a characteristic change in operating state became visible around four days before a documented failure. The state model was built in less than 15 minutes on an ordinary laptop.

No vast database of failure cases, no GPU farm, nor months of training a bespoke model.

And here is the point that matters.

The very same software core that analysed the pump also classifies radar signals.

Seventy-six target classes. F1 scores above 0.98 for almost all of them.

Completely different data. A Different physical domain. An entirely different task.

Yet no new software core.

How far does this principle go?

The SF2 team has put it to the test. Again and again.

A compressor. A plastics extrusion line. An offshore pipeline. A server farm. Tool wear monitoring. A conveyor track in semiconductor manufacturing.

Six completely different worlds. The same software every time.

That leaves one question: what happens when the subject is no longer a single machine, but an entire plant?

Three Functions. One Model.

The answer comes from a biogas plant in northern Lower Austria. Three very different plant functions operate side by side. A combined heat and power unit generates electricity and heat. An agitator keeps the biological process running. A gas flare burns off gas when it cannot be utilised. SF2 captured all three in a single state model using approximately 70 days of operating data.

Then the timeline was rewound.

The historical analysis revealed early precursors: 62 hours before a protective alert on the agitator, 53 hours before a gas-flare malfunction and, for the combined heat and power unit, a full 14 days ? two whole weeks ? before a major sequence of faults.

All results stem from documented, partially retrospective evaluations. While not a blanket guarantee for every installation, they prove a clear principle: the same software core applied to nine entirely different use cases.

Nobody Programs a New Spreadsheet Program for Every Balance Sheet.

A spreadsheet application is built once, and millions of people use it to run calculations with their own numbers.

In industrial AI, the opposite often happens. A bespoke solution is developed for every machine: collecting data, hiring specialists, training models, integrating. When the next machine class comes along, much of the process starts again.

For one machine, that bespoke approach works. For a thousand, it becomes unaffordable. SF2 reverses this principle: the technology stays the same; the user defines the application.

Data, preprocessing and reference states change but the core remains. And the models are built by those who know their equipment best: the machine experts themselves.

Not one model for every machine, but one technology that lets users build their own models ? just like in a spreadsheet.

Testing Is Explicitly Encouraged

And nobody needs to wait for it.

The SF2 Suite SE is permanently available free of charge. Users can analyse historical data locally, build their own state models and compare fingerprints without additional hardware and without data ever leaving the company. While the Suite RT extends this to live data, the Engine RT runs directly at the machine.

How small can ?directly at the machine? be?

On an NXP i.MX91, SF2 processes pump data in real-time replay and supplies state information for SCADA alarms.

On a 15-metre conveyor test track, SF2 identified defective sections using a single sensor-equipped container travelling along the line instead of installing sensors on every roller. Subsequently, the software was implemented on an Infineon PSoC Edge E84 AI Kit as part of the Thin[gk]athon organised by Infineon Technologies and ZEISS Digital Innovation.

For machinery manufacturers and OEMs, state intelligence thus becomes a software feature that can be embedded in existing devices, controllers and architectures.

?Companies don?t ask us whether SF2 is familiar with their specific machine,? says Christoph Gretzmacher, Business Development at SF2 Systems. ?They ask how quickly they can get started with their own data. The answer is: today. And if someone needs more ? such as a new interface or a specialised integration ? we build that too. The core stays the same; the solution grows with the customer.

Back to the Blood Test.

Every laboratory value is within its reference range.

But does that necessarily mean the person is healthy?

A single value can be completely normal and still be far removed from what was normal for that particular person. Several values can shift together without any single one triggering an alarm.

What if we lined up one person?s blood tests over several years, like the frames of a film?

The question would no longer be just: Are my results normal today? It would also be: Where is my overall biological state heading?

That is precisely what SF2 Systems is investigating in its publicly announced research collaboration with the Department of Laboratory Medicine of the Medical University of Vienna at Vienna General Hospital (AKH): Can routine clinical data yield Patient State Fingerprints that reveal changes in biological equilibrium ? homeostasis ? potentially before individual values cross diagnostic thresholds?

This is research, not diagnosis.

Whether such patterns are clinically meaningful remains to be established. SF2 is not an approved diagnostic procedure.

But consider what is happening here:

The same software that characterises a pump and distinguishes radar targets is now being explored on human laboratory data.

Not because humans are machines, but because the underlying question is the same: Where does a system stand ? and where is it heading?

The Real Breakthrough Isn?t the Ninth Application.

It is that it didn?t take a ninth technology to achieve it.

Every new use case expands the scope of the exact same, off-the-shelf software core, without having to start from scratch.

Collecting data gets easier every day. Understanding it does not.

Not a new AI for every problem ? just one core for many.

More information: https://sf2systems.com

Software: https://sf2systems.com/en/produkte/software/index.html

Technical case studies: https://sf2systems.com/case-studies/index.html

SF2 Systems GmbH

SF2 Systems is a Vienna-based deep-tech company specializing in software-based condition analysis of complex machines, industrial assets, and processes across industry, energy, and critical infrastructure. Vendor- and sensor-agnostic, the technology combines existing operational data into comparable state fingerprints. This makes drift and gradual degradation visible early ? reproducibly, transparently, without black-box models, and without requiring large labeled failure datasets.

Designed to complement existing condition-monitoring and automation systems, SF2 is built for brownfield and OT environments. It runs locally at the edge or on-premises with zero cloud dependency, keeping sensitive operational data strictly in-house. This enables early anomaly detection, reduces unplanned downtime, and makes maintenance far more predictable.

Additionally, SF2 is exploring applications for its core technology in biotechnology and life sciences.