Bearings and motors
A motor's vibrations are compared with its normal signature: when it changes, maintenance is planned before the failure.
The failure announces itself. You just need to know how to listen.
Vibrations, temperatures and work cycles become forecasts: maintenance arrives before the line stops. Statistical process control and forecasts work on the same stream of measurements.
Processes: the failure announces itself.
Vibrations, temperatures and work cycles become forecasts. A signal that goes out of limits becomes an alert, not a surprise.
Vibrations and work cycles reveal the tooth that is wearing out, before the line stops.
Measurements are sampled continuously and compared with the control limits.
Low-cost sensors collect the measurements and send them along the cable to the computing module.
When a value goes out of range, the alert goes off and the line slows down before breaking.
Why it matters.
A line stoppage costs more than the maintenance that would have prevented it. Many failures announce themselves weeks ahead: a vibration that changes, a temperature that drifts, a cycle that lengthens. The problem is collecting these traces and reading them in time.
Statistical process control and predictive maintenance work on the same data: the former checks whether the process stays within limits, the latter estimates when a component will leave its normal state.
How it works, layer by layer.
From measurement to intervention, in five layers.
- Sensors
- Accelerometers, temperature probes and cycle counters mounted on the machine.
- Edge
- A module close to the machine computes indicators such as RMS value and spectrum, and sends only what is needed.
- Models
- Control charts and diagnostic models compare measurements with the normal state and estimate remaining life.
- Alerts
- When an indicator leaves its limits, an alert is raised with the machine, the measurement and the threshold exceeded.
- Integration
- The information reaches ERP, SCADA and maintenance systems, so the intervention enters the work plan.
From signal to forecast.
Statistical process control and predictive maintenance work on the same stream of measurements: only the question you ask changes.
- Predictive maintenance
- Diagnostic models on machines and lines, from the sensor to the edge up to the cloud.
- Process control
- Control charts on the stream of measurements and alerts when a value goes out of limits.
- On-machine sensors
- Low-cost sensors with accelerometer and gyroscope, and lightweight models that run close to the machine.
- Backend
- FastAPI and PostgreSQL with a schema per client and monthly data partitions.
Use cases.
Statistical control on a line
Process measurements, such as thicknesses or temperatures, feed control charts that flag drifts before they leave tolerances.
Computer vision for quality control
Defect detection on high-speed surfaces, integrated directly into the existing PLC. No operator, zero line interruptions, documented scrap reduction.
Data, integrations and standards.
We connect to PLCs, SCADA and ERP with standard industrial protocols, for example MQTT, OPC UA and Modbus. Measurements are stored in monthly partitioned tables, and models can run at the edge or on the platform.
Privacy, security and compliance.
- Industrial security
- Data leaves the production network only through authenticated, encrypted channels, following good practice for industrial systems.
- No personal data
- Measurements concern machines and processes, not people: if an operator appears in the data, its use is defined beforehand.
- AI Act
- Models are documented, with training data, limits and performance, and their classification under the AI Act is checked case by case.
How to start.
- 01
Analysis and scope
We choose the critical machines, the failures to anticipate and the data already available.
- 02
Pilot on a real case
We mount sensors on one machine and collect its normal behaviour.
- 03
Extension
We add machines and lines, and refine thresholds and models with the history.
- 04
Operation and support
We monitor model quality and update the models when the machines change.
Words worth knowing.
- Predictive maintenance
- Interventions decided from the machine's actual condition, not from a fixed calendar.
- Statistical process control
- A method that follows a measurement over time and flags when it leaves normal variability.
- Remaining life
- The estimate of the time or cycles a component has left before intervention is needed.
- Vibration signature
- The typical pattern of a machine's vibrations in good condition, used as a reference.
Frequently asked questions.
How long does it take to get the first predictions?
The machine must first be observed under normal conditions, usually for a few weeks. Threshold-based alerts can start immediately.
Do we have to replace our SCADA?
No: the platform sits alongside existing systems and uses their data, without replacing them.
Is artificial intelligence really necessary?
No. Many failures are caught with statistical controls and well-chosen thresholds; learning models are added where they bring a measurable advantage.
The other areas.
People
Museums, kiosks and proximity marketing: tools that speak to people in the right place.
Open the pagePlaces
Indoor and outdoor localization, augmented reality search and contextual digital signage.
Open the pageTerritory
Multi-tenant GIS for public bodies: maps, open data and aggregated flows of people.
Open the pageSensors
IoT and Edge AI: models on the device and nodes that coordinate.
Open the pageSoftware
Custom software and cloud platforms, one sheet at a time.
Open the pageA line you would like to understand better?
Tell us which machines you have and which stoppages cost you the most.
Let's talk