R&D project in progress · Galician Industry 4.0

The plant already has the sensors. What it lacks is the diagnosis

Aura-Indus Q detects that an asset has begun to degrade before it shows up in the usual indicators, forecasts which other machines on the line will be affected and when, and explains it in the physical variables the maintenance engineer already knows.

Hybrid platform of predictive AI and quantum computing for orchestration and advanced maintenance in Galician Industry 4.0

What the project is, and who funds it How it works

Project funding

Aura-Indus Q is funded with the Xunta de Galicia’s own funds, through the Igape.

Proyecto financiado por la Xunta de Galicia a través del Instituto Galego de Promoción Económica (Igape), convocatoria IA360, procedimiento IG408M, Línea A.

IA360

Igape · procedure IG408M, Line A

Terms published in the Diario Oficial de Galicia, no. 10 of 16 January 2026
18

months, from March 2026

The industrial pilot in a real Galician plant starts on 17 March 2027
TRL 5 → 7/8

Base technology from ResilMesh

What predicted the lateral spread of an attack across an IT network now predicts a mechanical failure spreading across a production graph
The AuraSec project
Under construction

Specified, not yet measured

No performance figures on this page, because none has been measured yet

The problem

A second-tier supplier is instrumented, and still cannot say which asset is about to take the line down

The plant has SCADA with hundreds of signals per line, an MES, drives with the manufacturer’s own analytics and two or three years of history. What it does not have is an answer to the operational question: which asset is going to degrade before it degrades the line, and what it takes down with it when it does.

01

The useful signal lives in pre-failure, where it cannot be seen

By the time vibration spikes it is already late. The signature of deterioration — sideband modulation in a bearing, a few milliseconds of drift between two servos, a new harmonic in the stator current — has an amplitude comparable to process noise.
02

The tools are per asset, and the line is a graph

A compressor degrading does not show up as a compressor alarm: it shows up as a rise in rejects three stations downstream. Composing that across the line lives today in the heads of two or three people with twenty years on the floor.
03

A number without an argument gets switched off

A system that says “probability of failure 0.87” without saying why is unplugged within three months. The operator cannot act on a probability, and the person responsible cannot stop a line on a figure that comes with no reasoning attached.

What it does

A non-invasive layer of intelligence over the sensors the plant already has

It connects to the existing sensors and SCADA without replacing them and without touching the machines’ programming. It warns earlier, it says who the degradation is going to reach, and it explains itself in the maintenance engineer’s own language.

01

Listens, never writes

A DIN-rail node in the electrical cabinet subscribes to the OPC-UA, Modbus or MQTT the plant already exposes, with a credential that has no write permission. It installs without stopping the line.
02

Scores at the edge, with no cloud and no link

The anomaly score is deterministic and computed on the node itself. If the link drops, the node keeps scoring and holds up to sixty days. The data does not leave the plant in order to be scored.
03

Propagates across the line graph

Over the ISA-95 topology that process engineering maintains, it forecasts which assets are affected and on what horizon. The graph is declared by the plant, not guessed by us.
04

Explains, and leaves it on the record

The explanation is projected back onto the physical variables and presented in four levels: the three-second glance from eight metres away, the actionable notice in thirty seconds on a tablet and with a glove on, the evidence to decide with, and the full exportable record. All of it chained and signed before anyone sees it.

Three things worth saying plainly, because almost nobody says them: the data does not leave the plant in order to be scored; the only thing the system ever writes into a customer system is a work order proposed in the CMMS, in proposed state, with a credential that can neither execute nor close it; and if the chain of explanation cannot be produced, the system fails closed and marks the case as not decidable instead of inventing a reason.

Objectives and results

What the project has to have achieved by September 2027

Five objectives, each one with the result that proves it and a date on which it can be checked. This block is updated as the project delivers, and the results are published whichever way they come out.

Detect degradation in pre-failure

Result
A detection engine validated against signals whose amplitude is comparable to process noise, with the method recorded

Verifiable from
M6 · 09/2026
Not yet
Forecast the spread across the line

Result
Propagation over the plant graph declared by process engineering, with a relative horizon of impact per asset

Verifiable from
M12 · 03/2027
Not yet
Explain every notice in physical variables

Result
A frozen, auditable explanation procedure that fails closed when the chain cannot be produced

Verifiable from
M12 · 03/2027
Not yet
Validate it in a real Galician plant

Result
An industrial pilot on a production line, taking the technology from TRL 5 to TRL 7/8

Verifiable from
M18 · 09/2027
Not yet
Release what can be released

Result
Peripheral components under Apache-2.0, synthetic datasets on Zenodo with a DOI, a whitepaper and a pre-print

Verifiable from
M18 · 09/2027
Not yet

Nothing in this column is measured yet. It is filled in as each deliverable closes, and the figures only appear here once there is an instrument behind them.

Who it is for

Four sectors of Galician industry, and a different reason in each one

Naval and auxiliary

Propulsion, gearboxes, generator sets

The noise of swell and cavitation is non-stationary: “normal” is a set of regimes, not one.
Automotive

Presses, welding cells, synchronised servos

The failure that matters is the loss of synchrony between servos, not the breakdown of one of them, and OEE is lost in micro-stoppages the MES never classifies.
Wood, stone and forestry

Saws, planers, abrasive cutting lines

Dust blinds the sensors: the sensor degrades at the same time as the asset, and the two degradations get confused with each other.
Canning and food

Autoclaves, dosing and seaming lines

The real cost is not the stoppage but the loss of the batch, which is discovered hours after the event that caused it.

What it will not do

It does not close the loop, and that is a design decision

It does not write set points, stop lines or change process parameters

It proposes; the person decides and the plant executes. The subscription to the plant’s buses carries a credential with no write permission, and the operator’s decision is recorded in a way nobody can alter afterwards.
It does not replace functional safety

It is deployed alongside the existing safety functions, never in place of them, and it is not in the path of any of them.
It does not promise remaining useful life

It does not estimate hours of life left. What it delivers is detection of a change of regime, a forecast of propagation across the line, and a relative horizon of impact. Anything beyond that would be a number we cannot stand behind.
It does not claim regulatory conformity

It produces traceable, reproducible technical evidence on which a conformity assessment will have to rule. It is built to the high-risk requirements by our own decision, and we say it that way rather than assuming a conformity nobody has ruled on yet.

The project

What is being built, over what period, and with whose money

Aura-Indus Q is an eighteen-month industrial research and development project run by Tesla Technologies & Software S.L. from its R&D centre in Santiago de Compostela. The aim is to take a technology validated at TRL 5 in a European cybersecurity project and bring it to TRL 7/8 in a real Galician plant, in the form of a subscription service per monitored asset.

Holder: Tesla Technologies & Software S.L. · NIF B70305578 · Rúa Fontiñas 92, 15703 Santiago de Compostela
Call: IA360, aid for innovation and artificial intelligence · procedure IG408M, Line A · Instituto Galego de Promoción Económica (Igape), Xunta de Galicia
Technology origin: The base technology (TRL 5) comes from the ResilMesh project, European Union Horizon Europe programme, Grant Agreement no. 101119681
Period: 18 March 2026 – 17 September 2027 · industrial pilot from 17 March 2027
Alignment: RIS3 Galicia, Challenge 2 · Digitalisation and Sustainability priorities

Follow the project

There is nothing to try yet, and that is the honest answer

The industrial pilot starts in March 2027 and the results are published in June of that year, whichever way they come out. Until then there are two useful things to do: get on the list for the publications, or tell us about your plant if you want to be considered for the pilot.

A plant interested in the pilot Our other R&D projects

Proyecto financiado por la Xunta de Galicia a través del Instituto Galego de Promoción Económica (Igape), convocatoria IA360, procedimiento IG408M, Línea A.