Home / How we work
Methodology
Five steps, from measurement to operational decision
The methodology applies consistently to any context: industrial, agricultural or domestic. What distinguishes each project are the sensors installed and the parameters defined with the team. This structured approach is what underpins, in practice, an Industry 4.0 strategy.
Measure
We install wireless sensors at the points that determine the outcome of the operation. The wireless link allows installation while production continues, and most sensors stay operational for years on a single battery.
The first technical site visit determines two critical factors: which quantities determine the outcome of the operation, and whether the radio signal reaches the points where those quantities are measured. We bring a gateway and a set of sensors to validate link quality at the relevant points, including basements, plant rooms and steel-framed buildings, which represent the most demanding scenarios.
We then select the right sensor for each measuring point. Installing a new sensor is not always necessary: where a meter, a PLC or a drive already exists, we integrate a reading from the existing equipment instead of duplicating the measurement. This approach lowers the investment and avoids discrepancies between two sources for the same quantity.
Reporting frequency is defined case by case, according to the criticality of each point. A cold room requires readings every few minutes; a tank level can be read every fifteen. Reading less often extends battery life, so we calibrate the frequency to the minimum each application requires.
Monitor
Readings arrive at our servers and are stored with their timestamp. From there, we build the monitoring views around each client's real process, rather than around a generic template.
The screens are designed around each client's specific process. We use the names the team already uses for the equipment and the units they are used to working with. A dashboard that forces people to translate code names in their head stops being opened after two weeks.
Each team gets its own view and its own access. Production needs the current state; maintenance needs the trend over recent weeks; management needs the monthly total. It is the same history, read in different ways, inside a single system.
The history is kept with the timestamp of every reading and stays available months later. That capability is what lets you compare two shifts or two months without depending on what anyone remembers.
Model
With a few weeks of history, we build the reference for what is normal on each site: the consumption expected at a given production rate, the usual curve of a tank, the signature of a machine in good condition.
A single value carries little information. Forty degrees in a tank can be normal in mid-afternoon and signal a problem at six in the morning. The reference is built from the history of the installation itself, not from catalogue figures, because every site has its own rhythm, its own hours and its own environmental conditions.
From this reference, we compare each reading with what would be expected at that moment. It is this comparison that gives the number its meaning, and that distinguishes ordinary seasonal variation from equipment that has started to degrade.
Modelling takes time to mature. In the first weeks there is only measurement and history; the reference forms as the system observes the installation under different conditions. We communicate transparently what can be estimated at each stage, without promising forecasts before there is data to support them.
Alert
We define, together, the thresholds for each measuring point. When a value leaves the agreed band, the alert is routed to whoever is on duty at that moment, naming the point concerned.
Thresholds are calibrated with the people who operate the installation, not inherited from factory defaults. A badly calibrated limit produces an excessive volume of alerts, and a system that flags everything stops being read. We start with wide bands and tighten them with the experience of the first weeks.
The alert is routed to whoever is on duty at that moment, by e-mail, SMS or mobile notification. Routing follows the shift and the duty roster, so that an event at three in the morning does not wait for office hours.
Every alert is logged, with the time it was raised and the outcome recorded afterwards. This record makes it possible to tell a one-off incident from a recurring problem, and serves as evidence when an operational decision needs to be justified.
Decide
With the history consolidated, decisions stop depending on memory. It becomes possible to compare periods, justify an investment and later confirm whether it produced the expected effect.
With months of accumulated history, there is no longer any need to reconstruct what happened. A rise in consumption has a date, a duration and a piece of equipment attached to it, which significantly shortens diagnosis time and leaves room to deal with the cause.
After an intervention, the same data confirms whether the effect held. It is common for consumption to drop the month after a change and return to its previous level a quarter later, without anyone noticing the reversal. With continuous measurement, that reversal becomes visible in the history.
Reports are generated by shift, line, vehicle or plot depending on the context, and can be exported for anyone who prefers to work with the data independently. The data belongs to the client and leaves the system at any time.
Start with a single measuring point
There is no need to instrument the whole operation at once. A small perimeter is enough to validate network coverage and confirm the value of the data in your specific context.