Know which sensor data from your machines you can trust.
A reliability score for every sensor channel in your installed fleet, with a readable cause code behind every flag. Software only: no new hardware, and no change to your machines.
Four problems that almost every connected fleet has
Four symptoms, one common cause.
False service alarms
Engineers travel for nothing, and the service department gradually learns to ignore the alarms, which quietly removes the value of the true ones as well.
Invisible sensor drift
Sensors in the field degrade slowly. The data stays plausible, which is exactly the problem, because plausible and correct are not the same thing.
Warranty disputes
Your machine failed, versus your sensor failed. Weeks of argument about what the data actually proves, with no neutral party in the room.
Predictive maintenance stalls
The service ambition is there, but nobody wants to build a commitment on data they are not sure of.
Often it is your sensor, not your machine
A vibration peak in the data can mean two things. A real machine problem, such as bearing wear, imbalance or misalignment, which needs action. Or a sensor problem, such as a loose mount, drift or saturation, which is a false alarm. Without context, the two look exactly the same in the data.
Only physics-aware analysis, which tests a measurement against speed, load and the behaviour of the sensors around it, can tell them apart.
60 to 80 percent
The share of a data team's time that typically goes on cleaning sensor data by hand before any analysis can start.
90 percent
The share of industrial sensor data that is never used at all, in part because nobody is confident enough in it to build on it.
The second figure is IBM's, and it matches what we find. The first is the range we see across the projects we are brought into.
Three layers produce one score
Statistical checks. Signal energy, variance, spectral stability and outliers. This catches the obvious faults, and it is where most tools stop.
Physics-aware consistency. Sensors are tested against each other and against operating context such as speed and load. This is the layer that separates a genuine machine problem from a sensor problem, and it is the reason the score can be trusted.
Drift detection. Machine learning finds the slow degradation that a fixed threshold will never see, because the value never leaves the plausible range.
Every flag comes with a readable cause code, so an engineer can see why a channel was marked rather than being asked to accept it.
So that every alarm carries an answer to the only question that matters first: is this channel usable, degraded or invalid?
Illustrative example of the output format, not measured data from a customer fleet.
Not a new product to sell. A harder foundation under what you already sell.
Service margin back
Fewer unnecessary engineer visits. Every prevented false alarm is margin that was going to walk out of the door, and the effect is immediate rather than strategic.
Stronger in warranty discussions
A neutral, explainable judgement on the data. Evidence in front of your customer instead of a disagreement about whose fault the reading is.
A foundation under servitization
Predictive maintenance and uptime guarantees can only be sold once the data layer underneath them is reliable. This is that layer.
It fits what you already have
We read the data you collect today. REST API, MQTT or OPC-UA. Running at the edge, in the cloud, or both. There is no hardware to buy and no rip and replace, and either historical data or a live stream is enough to start.
Your machines
The sensors that are already in them. Nothing is added and nothing is modified.
Your data
History or a live stream. An export is enough for a first assessment.
Tarucca Edge
A trust score from nought to a hundred per channel, plus the cause behind it.
Your service platform
Alarms and dashboards you can act on, inside the tools your people already use.
Built in the hardest sensor environment there is
Offshore wind: storm loading, constant vibration, assets nobody can reach, and decisions worth millions taken on the strength of sensor data. If a measurement chain holds up there, a factory floor is a gentler problem.
- Where we operate
- Eindhoven, on the TU Eindhoven campus, with AI, sensing and structural dynamics as the core and roots in the Brainport region.
- What we have worked on
- Sensor and data analysis on wind turbine blades, subsea cables and mooring systems, in collaborations including Eneco, Vattenfall, Shell, TNO, TU Delft and TU Eindhoven.
- The models
- Physics-aware, trained on vibration, strain and load data. Which is to say: exactly the kinds of sensor that are already fitted in your machines.
The Sensor Trust Scan
A small, bounded first step with a fixed scope and a fixed duration. No integration project, and no commitment beyond it.
A reliability score per sensor channel
Across the machines in scope, with the cause behind every score that is not clean.
A false alarm analysis
Which of your recent alarms were sensor faults rather than machine faults, and why.
A data quality baseline
For your installed fleet, so you know where you actually stand before deciding anything.
Recommendations and a debrief
A written report and a session with your management and engineering people together.
- What we need from you
- One data export covering at least three machines and at least six months of history, plus two sessions of an hour.
- Confidentiality
- A non-disclosure agreement as standard. Your data stays yours and is not used for anything else.
- After the scan
- If it is useful, the next steps are a pilot connecting live trust scores to your service platform, and then continuous monitoring across the fleet. Each step is a separate decision, and there is no lock-in: the report and the insight are yours either way.
How many of your alarms are actually sensor faults?
Most machine builders have a rough idea and no evidence. The first step is small: one data export, two conversations, and a written answer to that question for your own installed fleet. If the answer turns out to be very few, that is worth knowing too.
