¿Con qué datos se entrenan los modelos?

Con los de cada instalación: imágenes y señales del propio entorno, etiquetadas con casos reales. Un modelo genérico no conoce las condiciones de su planta.

¿Las imágenes salen de la instalación?

No es necesario. La inferencia puede ejecutarse en el propio equipo y enviar solo el evento detectado, sin transmitir las imágenes a la nube.

¿Qué es el mantenimiento predictivo?

Detectar en la vibración, la temperatura o el consumo de una máquina los cambios que preceden a una avería, para intervenir antes de que se produzca la parada.

COMPUTER VISION · PREDICTIVE MAINTENANCE

Artificial intelligence and detection

Computer vision and sensing to anticipate failures, incidents and intrusions.

In-house development

Request a technical meeting Tell us about your project

Early pest detection
Computer vision in the plant
Predictive maintenance
Perimeter and access control

Early pest detection

Computer vision in the crop

Models trained with images of the crop itself identify pests such as Tuta absoluta before they are visible to the naked eye.

Continuous monitoring, 24 hours a day, 365 days a year
Targeted treatments and less use of plant protection products
Local inference, without sending images to the cloud

Request a technical meeting →

Computer vision in the plant

Visual supervision of processes and people

Our own models to detect manufacturing defects, use of personal protective equipment or risk situations in the plant.

Models trained with images of the plant itself
Processing on the device, no images leave it
Retraining with new cases

Request a technical meeting →

Predictive maintenance

Failures anticipated

Neural networks that learn the normal vibration and consumption of each machine and warn before the failure occurs.

Detection of process anomalies
Planned maintenance instead of unexpected stoppages
Model retrainable with new cases

Request a technical meeting →

Perimeter and access control

Supervised perimeters and access points

Our own sensing, vision and communications to monitor perimeters, access points and restricted areas.

Intrusion detection at fences and access points
Event correlation with images
Real-time alert to the control centre

Request a technical meeting →

Software · Machine learning

Tailor-made i-Books: a neural network trained for your problem, not a generic model

An off-the-shelf model recognises what it was taught. Yours has to recognise what happens at your installation: your species, your part, your defect, your lighting. We design and train the model with your own data and deliver it packaged as an i-Book ready to run inside the device.

The value is not in the algorithm, it is in your data

Algorithms are public and everyone uses them today. What nobody else has is your history: how a pest looks in your greenhouse under your light, what a faulty weld looks like on your line, how your pump sounds before it breaks down. We start from that material, label it with your team and train a model that only works for you — and that is precisely why it works.

Your own modeltrained with your images and your history, not off the shelf

Local inferencedecides inside the device, without sending images to the cloud

Retrainableimproves every season with the new cases that appear

Yoursthe data and the resulting model belong to your company

Image recognition

Identifying pest species, manufacturing defects, PPE use, number plates, crop condition or any visual pattern that today depends on someone looking.

Purpose-specific neural networks

Failure prediction from vibration or consumption, harvest estimation, process anomaly detection or classification of events in your sensor history.

Delivered as an i-Book

The model doesn't stay in a lab notebook: it is packaged with its documentation, its metrics and its retraining procedure, ready to install on the device.

How an i-Book is built

A short, well-scoped project, with a success criterion agreed before starting.

01 · DEFINE

What needs to be decided

Before the model, the decision: what you want to detect, with what margin of error it is useful and what happens when it is right and when it is wrong.

02 · GATHER

Your data

We collect and label the material with your team, who know how to tell a good case from a bad one. If there is no history, we generate it in the field.

03 · TRAIN

And validate properly

We train and measure against cases the model has not seen, with the metrics that matter in your process, not just overall accuracy.

04 · DEPLOY

Inside the device

It is optimised to run on the hardware, integrated into the ecosystem, and the retraining procedure for new cases is documented.

Tell us about your project

At the first meeting we review what data you have and tell you frankly whether your case needs machine learning or can be solved with a well-placed rule.

Photovoltaics · Prevention, alert and extinguishing

We don't sell panels: we protect the ones you already have

Our IoT ecosystem monitors photovoltaic installations already in place, on industrial roofs and on the ground, with one network of our own modules. It prevents before an incident occurs, alerts when one appears and extinguishes if it breaks out. A single solution that serves every panel and every string box equally.

Industrial solar roofs

A power plant on top of your business

On an industrial roof, what lies beneath the panels is your building: production, warehouse and people. There, an electrical incident is not just lost generation. Our modules are distributed across the rows and the string boxes, and monitor continuously without interfering with the operation of the installation.

Panel-by-panel monitoring and at each string box
Our own modules, with no works on the existing installation
Action on the affected point, not on the whole roof
Simultaneous alert to your team and the dashboard

Ground-mounted solar fields

Kilometres of rows that nobody walks every day

In a solar plant the problem is distance: many hours and many rows fit between two maintenance rounds. The ecosystem covers the full area with the same network of modules, so that every row and every string box is looked after even when nobody is nearby.

Coverage of the whole plant, row by row
Our own industrial communications up to the dashboard
Weather station integrated into the same network
Local response: it doesn't depend on someone watching

An ecosystem that prevents, alerts and extinguishes

A sensing layer is distributed across the installation over a robust industrial network, with complete isolation between our system and the photovoltaic one. Everything is supervised from the client's control panel, which shows the plant status, environmental telemetry and any deviation from reference values. The ecosystem can work in intelligent automatic, supervised or manual mode.

The client supervises the entire installation from their own control panel: status of each zone, environmental telemetry and any deviation from reference values.

Intelligent automaticThe ecosystem decides and acts on its own on the affected point.

SupervisedIt proposes the action and waits for your team's confirmation.

ManualControl remains entirely in the operator's hands.

PreventsA discreet extension installed where it is needed: on the solar panel, in the string box or in any electrical box. It monitors continuously from day one.

AlertsOver a robust industrial network, completely isolated from the photovoltaic installation. It triggers on sparks, light or arcs, the presence of chemicals, smoke or flame, and the deviation is instantly shown on the dashboard.

ExtinguishesOnce fire is confirmed, the patented and approved Extincise extinguishing acts on the affected point. The plant stays up: the incident is put out, not the installation.

No third partiesno external dependencies

No external APIsoperates as a closed system

No paid appsno fees or licences

Your data, in your systemhosted at your own plant

Complete isolationseparate from the PV system

Robust industrial networkdesigned to operate 24/7

Three operating modesautomatic, supervised, manual

Environmental telemetrytemperature, wind, humidity, irradiance

Decides on sitelocal response, without waiting for anyone

Full coveragepanel, string box and electrical box

European regulatory framework for the sector

EN 54 — Fire detection and fire alarm systems
EN IEC 61730 — Photovoltaic module safety
EN IEC 62446 — Testing, documentation and maintenance of PV installations
Regulation (EU) 305/2011 — Construction products (CPR)
Directive 2014/30/EU — Electromagnetic compatibility
Directive 2014/35/EU — Low voltage

Our modules are designed in accordance with EN 54. These are the reference standards applicable to the sector; the specific regulatory scope of each installation is determined in the technical phase of the project.

Request information

360° IoT SOLUTION

A real case of a business problem in the primary sector

IoT ecosystems that monitor crops 24h · 365 days. Predictive and preventive system.

AUTOMATIC PEST RECOGNITION

Early pest detection using computer vision

In the primary sector, pests are the silent enemy: when they are spotted with the naked eye, the damage is already done. Our computer vision systems trained with machine learning continuously analyse crop images and identify the presence of insects and the first signs of infestation before they are visible to the human eye, turning each farm into an asset monitored 24 hours a day.

The case illustrating this section —the tomato leafminer (Tuta absoluta)— is one of the most costly in the world. Early detection makes it possible to intervene in a targeted way and drastically reduce the use of plant protection products, protecting the grower's margin and the sustainability of the farm.

20–40%

of global agricultural production is lost to pests every year

up to 100%

tomato crop loss can be caused by Tuta absoluta

80–95%

accuracy in automatic pest identification

24/7

continuous crop monitoring, without relying on manual inspection

01 · Capture

Continuous images of the crop in the field and greenhouse.

02 · Analysis

Automatic recognition of insects and early signs of damage.

03 · Alert

Immediate notice to the person in charge, with the location of the incident.

04 · Decision

Targeted treatment: less plant protection product, higher yield.

Discover our engineering

Frequently asked questions

What data are the models trained with?

With those from each installation: images and signals from the environment itself, labelled with real cases. A generic model does not know the conditions of your plant.

Do the images leave the installation?

That isn't necessary. Inference can run on the device itself and send only the detected event, without transmitting the images to the cloud.

What is predictive maintenance?

Detecting in a machine's vibration, temperature or consumption the changes that precede a failure, so as to intervene before the stoppage occurs.