Predictive Maintenance in Emilia's Industrial Districts: Where to Start

In Emilia’s industrial districts, unplanned downtime has a price that everyone knows and almost nobody measures: a ceramic press stopped in Sassuolo, a cartoning machine down in the packaging valley near Bologna, a machining line halted in the Motor Valley. Depending on the line, that means thousands or tens of thousands of euros per hour of downtime — plus the penalty on late delivery, plus the batch that has to be reworked. Predictive maintenance exists to avoid exactly this. But between saying and doing there are three things that generic articles never explain: what “predictive” actually means, which technical standards to start from, and how it gets paid for. Here we set out all three, in order.

“Predictive” is a precise word (and it matters in tender specifications)

The UNI EN 13306 standard — Europe’s maintenance vocabulary — distinguishes three things that are often confused in quotations:

Cyclical preventive maintenance: intervention on a calendar basis, every N hours or N cycles, whether it’s needed or not. Simple, but you pay for the replacement part even when it wasn’t due.
Condition-based maintenance: intervention when a measured parameter crosses a threshold. Already better: the part is replaced once it starts to degrade.
Predictive maintenance: from the trend in the parameters, I estimate when degradation will turn into failure, and I schedule the intervention ahead of that date, within a convenient production window.

The difference is not academic: when you sign a contract for a “predictive maintenance system”, the EN 13306 definition is what entitles you to demand a forecast with a time horizon — not a traffic light that turns red once vibration is already off the scale. Ask your supplier which of the three they are actually selling. The answer is instructive.

The data is already there: the first sensor is the PLC

The most costly mistake we see is starting by buying sensors. In Emilian production departments — hydraulic presses, roller kilns, spray dryers, filling machines, machining centres — the PLCs and SCADA systems already log electrical current draw, pressures, temperatures, cycle times, alarms. This is data the company already owns and almost never uses: in most cases it gets overwritten, or sits in historian databases that nobody ever queries.

Before installing a single accelerometer, the right question is: what does the data I already have tell me? A current draw that rises for the same recipe means increasing friction. A cycle time that lengthens means mechanical or pneumatic degradation. A temperature drift on a kiln is a refractory or combustion problem announcing itself weeks in advance. Bringing this data together into a single model of the plant — it’s the first step of every one of our operational trials — costs a fraction of instrumentation and often covers the first use cases on its own.

Dedicated sensors (vibration first and foremost) come later, on the assets where the PLC isn’t enough: spindle and fan bearings, gearboxes, pumps.

The two standards that save you months

When you move on to vibration, there’s no need to invent thresholds by gut feeling. Two technical references exist that almost no popular article ever cites, and they are worth their weight in gold:

ISO 17359 is the general condition-monitoring procedure: how to choose what to monitor (criticality × probability of failure), which parameters to observe for each machine family, how to set up the programme. It’s the checklist that keeps you from ending up with a toy project monitoring the wrong machine.
ISO 20816 (which superseded the long-standing ISO 10816) defines the vibration severity zones — from Zone A (new machine) to Zone D (vibration severe enough to cause damage) — with reference values for machine classes. The first-alarm thresholds, pending construction of your own plant’s specific baseline, are taken from there — not from the sensor supplier’s manual, which has every incentive to make as many alarms sound as possible.

The operational point: ISO thresholds are the starting point, the baseline per operating regime is the destination. A press working thick porcelain stoneware vibrates differently from the same press on a thin format: a serious system learns the signature of each regime and flags a deviation from its own normal, not the crossing of a fixed number. That’s the difference between three false alarms a week — which the department learns to ignore — and an occasional, well-justified alert with the trend attached.

The business case, done honestly

The ROI of predictive maintenance is worked out in three lines, and it pays to do it before signing anything:

Hourly cost of downtime on the candidate line: lost hourly margin + hard costs (idle staff, restart scrap, any penalties).
Hours of unplanned downtime per year on that line, from the maintenance history (if no history exists, that’s the first problem to fix — and it costs nothing).
Avoidable share: the industry literature indicates that a substantial proportion of mechanical failures give measurable signals weeks in advance; even prudently assuming you catch only half of them, on a critical line the sums almost always work out.

If the figure at point 1 is high and point 2 isn’t close to zero, predictive maintenance pays for itself on that single department alone. If, on the other hand, the line rarely stops and costs little when it does, it’s better to invest elsewhere: predictive maintenance is not a moral obligation, it’s an economic decision.

The 2026 news: hyper-depreciation now covers software too

Here’s the news that many haven’t yet registered. The 2026 Budget Law (Italian Law 199/2025) has retired the Transition 4.0 and 5.0 tax credits and reintroduced hyper-depreciation: for investments made from 1 January 2026 (with a window running to 30 September 2028), the cost of qualifying 4.0 capital goods is deducted at an increased rate, on a tiered scale — full uplift on the portion up to €2.5 million, tapering above that.

Three points that matter for anyone assessing a predictive maintenance project:

Software qualifies among the eligible assets: monitoring and analytics platforms, if capitalised and interconnected, benefit from the uplift. Under the old tax credits, the software component was the poor relation; not any more.
Interconnection is a requirement: the asset must communicate with the production management system. That is exactly the data-integration work described above — the tax compliance step and the technical project coincide.
The usual compliance conditions remain (workplace safety, social security contributions), along with the ban on stacking other aid on the same costs.

The implementing decree sets out the forms and filings: before structuring the investment, a conversation with your tax adviser about the exact configuration is essential — MIMIT’s reference page is the one on the new plan. But the substance is clear: in 2026, a well-built data-plus-monitoring project costs, net of the tax benefit, significantly less than its list price.

Where to start, in practice

Choose a single line, the one where downtime costs the most: the calculation in the previous section tells you which.
Make the data you already have speak: PLC, SCADA, maintenance history. Two weeks of analysis tell you more than a year of meetings.
Use ISO 17359 to decide what to monitor and ISO 20816 for the initial thresholds: no gut feeling, no supplier-set thresholds.
Demand the baseline per regime and a clear process: who receives the alert, who decides, within what window the intervention is scheduled. An alarm without a process is just noise.
Align the project with the tax benefit together with your adviser before placing the order, not after.

This is the same path we follow in our operational trials in manufacturing: one line, the data that already exists, a use case that pays for itself — and you grow from there. And if your company also falls within the scope of NIS2, the data work is the same: a single inventory, two obligations resolved.

Want to find out whether your most critical line is a good fit? A 30-minute session with one of our experts is enough to work out the sums together.