Predictive maintenance
Bearing damage, imbalance and cavitation surface in the residual weeks before failure — reported with a cause, not just a threshold crossing.
Vibration · AcousticsIndustrial edge-AI × physics-based models
Physics already knows half of it.
We teach the machine the rest.
Noether builds hybrid models — known physical law as the prior, machine learning for the residual — and runs them in real time on edge hardware next to the machine. No cloud in the loop. No data hunger. Alarms you can explain.
- <10 ms
- inference at the machine — no cloud in the control loop
- 21.7%
- CAGR of the edge-AI market, 2026–2033
- 27%
- CAGR for predictive maintenance in Europe, 2026–2033
- 10–100×
- less training data when physics carries the structure
Method
Grey-box: physics where it holds, learning where it doesn't
Pure machine learning needs failure data that industry rarely has — breakdowns are expensive and therefore rare. Pure physics modelling misses the messy reality of wear. The hybrid takes the load-bearing half from each.
Physical law as the prior
Rotordynamics, heat transfer, fluid mechanics — what you already know about your process is encoded as model structure instead of relearned from zero. Conservation laws are built in, so predictions stay physically admissible even where data is thin.
ML learns the residual
The gap between model and measurement is small, structured and dense with information. That is where wear, imbalance and bearing defects live — and it can be learned from weeks of data rather than years.
The whole thing runs at the edge
Quantised models on an industrial NPU or MCU beside the machine. Millisecond latency, correct behaviour with the network down, and sensitive process data that never leaves the plant.
Applications
Where physics and data beat either one alone
Virtual sensors
Estimate quantities you cannot instrument — temperature inside the tool, torque along the shaft — from the sensors already fitted.
State estimationProcess optimisation
Trade energy against quality in real time under hard physical constraints, so the optimiser can never walk the process outside its safe envelope.
Energy · YieldExplainable anomaly detection
When physics is the reference, every deviation is interpretable: what departed, by how much, and from which conservation law.
Auditable by designEngagement
What a pilot looks like
Four weeks, one machine, a fixed fee. The point is not a slide deck — it is a model running on your own hardware against your own process, and an honest answer about whether the residual carries the signal we think it does.
- Week 1
One machine, one failure mode
We pick a single asset that costs you money when it stops, and write down the physics that governs it. No instrumentation project — we start from the sensors you already have.
- Weeks 2–3
Hybrid model on your data
The physical model is fitted to your process, and the residual model is trained on your historian. You see which deviations it can already account for and which it cannot.
- Week 4
Running on the edge box
The model is quantised and deployed on hardware at the machine, streaming live. You keep the box and the model whether or not we continue.
Why now
Three curves crossing at once
The hardware arrived
Industrial NPUs run quantised networks under ten watts. The edge-AI market grows from roughly $30bn in 2026 toward $119bn by 2033.
Pure ML hit its ceiling
Industrial datasets are too small and too expensive for data-hungry models. Physics-informed learning is where the research has converged — but few have turned it into product, and almost nobody has taken it to the edge.
The data cannot leave
Sovereignty requirements and IP protection push inference on-premise. European predictive maintenance grows 27% a year, and those buyers want it without a cloud dependency.
Sources: Research and Markets, Edge AI Market Report 2026; Grand View Research, Europe Predictive Maintenance Outlook. On the competitive picture — PhysicsX, SimScale and Monolith all sell cloud simulation into the design phase. Real-time operation at the edge is still open ground.
Founder
Built by a physicist who also writes the firmware
Noether sits where theoretical physics meets hardware-near programming: the differential equations, the signal processing and the code on the microcontroller from the same pair of hands. That combination is exactly what hybrid edge-AI demands, and it is rare — most teams have one half and outsource the other.
Anyone can train a network on cloud data. Far fewer can write down a process's Lagrangian, discretise it, and make it fit in 512 kB of RAM.
Next step
Do you have a machine that costs money when it stops?
We are looking for two or three pilot partners in Nordic manufacturing and process industry. You get a hybrid model running on your own machine; we get validation data from a real plant instead of a benchmark dataset.