Hyper-depreciation 2026–2028: predictive maintenance software is among the eligible assets. See under which conditions→
A guide for SMEs and large companies

Predictive maintenance: act before the failure, only when needed.

How it works, what data it needs, what it returns, and why it is within reach both for an SME with a few critical machines and for a group with hundreds of plants.

€69,000the value of a single avoided failure on a critical machine, in this guide's example
48 h → 8 hfrom a three-day unplanned stop to a job done during a planned stop
3–10critical assets to start from: you don't need to monitor everything
weeksto learn normal behaviour; the first useful anomalies within the first months

Figures from the illustrative example in the Return section: costs and times vary widely by sector and asset.

In short

Fewer unplanned stops and fewer unnecessary jobs, starting from the machines that cannot stop.

  • It uses data on the real condition of machines and plants (vibration, temperature, current draw, process parameters) to predict when a component will fail and act just before.
  • It comes after run-to-failure maintenance (repair when it breaks) and calendar-based preventive maintenance (replace on schedule, even if the part is still good).
  • The benefit is twofold: fewer unplanned stops and fewer unnecessary jobs on components still in good condition.
  • You don't need to monitor everything: start from critical assets, those whose stop blocks production or costs the most.
  • It is within reach of SMEs too thanks to low-cost sensors, wireless connectivity and cloud platforms. In large companies it lets you manage fleets of plants with uniform criteria.
  • It uses the same data as the MES and interconnected machines, and predictive maintenance systems may qualify for Italy's 2026–2028 hyper-depreciation.
Definition

Predictive maintenance is a strategy that estimates the health and remaining life of a component from continuous or periodic measurements of how it runs, and schedules the job when degradation reaches a critical threshold, before the failure.

The European standard UNI EN 13306 on maintenance terminology describes it as condition-based maintenance carried out following a forecast derived from the analysis and evaluation of significant parameters of the item's degradation.

Why we talk about it. Nexus connects machines, sensors and company systems and turns operating data into information for the people who decide on maintenance jobs. Predictive maintenance is where data collection, software and maintenance organisation must work together.

Strategies compared

Predictive maintenance doesn't replace the other strategies: it's used where it pays off.

Several approaches coexist in every plant, chosen asset by asset based on criticality and the cost of failure. Pick a strategy to see when it acts on the degradation curve.

When you act

Predictive

When the model forecasts the failure within a useful horizon

Advantages

Minimal unplanned stops, full use of component life, planned jobs

Limits

Requires sensors, historical data and analysis skills

Suited to

Critical, costly assets with no redundancy

The degradation curve

Health of a component over time

Degradation curvepredictive window100%0time →component healthplanned job

From emergency maintenance to maintenance planned on data.

The useful window

Predictive maintenance works between the moment degradation becomes detectable and the functional failure: early enough to plan the job with the spare part ready, during a planned stop, late enough to use the component's life. Weeks or months usually pass between the two.

Legal obligations: predictive maintenance doesn't replace them. Some checks have intervals set by law (fire protection, periodic equipment inspections, earthing systems). Predictive maintenance adds to these checks, it doesn't remove them.

How it works

From sensors to the job: a chain of five steps.

Every weak link reduces the value of the whole chain. Select a step to see what happens.

Step 1 / 5

Acquisition

Dedicated sensors (vibration, temperature, current, pressure, ultrasound) or data already in the machine's PLC and CNC.

  • Accelerometers
  • Temperature probes
  • Spindle loads
  • Axis current draw
  • Cycles and alarms

Every weak link reduces the value of the whole chain.

Sensors on motor, gearbox and pump, a gateway on the shop floor, a dashboard and a technician with the spare part ready.

Where the data runs

Edge on the shop floor

Fast processing, works even without a network, reduces the data to transmit.

Cloud or central platform

Long history, comparison across machines and plants, more complex models.

Machine builder

Many builders offer remote monitoring of their machines: useful, but the data must stay accessible to the company too.

The three levels of analysis

Level 1

Thresholds and rules

An alarm when a parameter exceeds a limit, for example vibration beyond ISO 10816/20816 values.

Data needed
A few weeks of measurements, standard or manufacturer references
For whom
First step, suitable for everyone
Level 2

Anomaly detection

A model learns normal behaviour and flags deviations, even with no failure history.

Data needed
A few months of normal operation
For whom
SMEs and large companies, on critical assets
Level 3

Remaining-life forecasting

Statistical or machine-learning models estimate how long until the failure.

Data needed
History of failures and jobs, many similar assets
For whom
Large machine fleets, machine builders
The practical point: you don't need to start with artificial intelligence. Well-chosen thresholds and anomaly detection cover most of the value; remaining-life models come when the history is rich enough.
Monitoring techniques

Every failure has its signal. Every signal has its technique.

Choose the type of machine to see which techniques apply and which failures they help anticipate.

Machine type

Vibration analysis

What it measuresVibration amplitude and spectrum

Typical failures detectedImbalance, misalignment, bearing and gear wear, looseness

Motors, pumps, fans, gearboxes, spindles

Thermography

What it measuresTemperature distribution

Typical failures detectedOverheating, loose electrical connections, friction, degraded insulation

Electrical panels, motors, furnaces, bearings

Motor current analysis

What it measuresCurrent draw and waveform

Typical failures detectedRotor and stator faults, overloads, downstream mechanical problems

Electric motors, compressors, conveyors

Oil analysis

What it measuresParticles, viscosity, water, contaminants

Typical failures detectedInternal wear, contamination, lubricant degradation

Gearboxes, hydraulic circuits, engines

Ultrasound

What it measuresHigh-frequency acoustic emissions

Typical failures detectedAir and gas leaks, electrical discharges, poorly lubricated bearings

Compressed-air networks, valves, bearings

Pressure and flow

What it measuresPressure and flow trends

Typical failures detectedClogged filters, leaks, pump and valve wear

Hydraulic and pneumatic systems

Machine controller data

What it measuresAxis loads, torques, cycles, alarms from the PLC or CNC

Typical failures detectedTool and guide wear, play, process drift

Machine tools, presses, robots

Product quality

What it measuresDimensional measurements, scrap

Typical failures detectedProcess drift that anticipates a failure

Lines with in-line inspection

Energy

What it measuresConsumption per machine and per cycle

Typical failures detectedIncreasing friction, leaks, operation off the optimal point

Compressors, motors, furnaces, air conditioning

The data you already have. Many modern machines, especially interconnected ones, already expose loads, current draw and alarms. Before buying sensors, check which signals can already be read from the controller: they are often enough to start.
The benefits

The failure becomes a planned job.

Fewer unplanned stops

The failure becomes a planned job.

Less unnecessary maintenance

Components are replaced when needed, not on schedule.

The right spares at the right time

Less safety stock, fewer urgent shipments.

Less knock-on damage

A bearing replaced in time does not ruin the shaft or the gearbox.

Better quality

Machine degradation often shows up first in scrap.

More safety

Fewer sudden breakdowns and fewer emergency jobs.

Less energy

A degrading machine uses more energy.

Knowledge protected

What the expert technician knew «by ear» becomes a recorded measurement.

SMEs and large companies compared

Same logic, different scale

Goal
Keep the critical machine running
Initial scope
1–5 critical assets
Technology
Wireless sensors, cloud, alarms
Analysis
Thresholds and anomalies
Who runs it
Owner, shop-floor manager, external technician
Return
The first avoided failure

In an SME one or two machines are often the bottleneck of the whole production, with no backup machine. A stop of a few days means missed deliveries.

  • Start small: a few wireless sensors on critical assets, a subscription cloud platform, thresholds and alarms on smartphone.
  • It makes up for not having a maintenance department: the system acts as a continuous «first observer», instead of inspection rounds nobody has time for.
  • External maintenance is under control: data lets you call the technician or the machine builder with a diagnosis ready, and evaluate service contracts.
  • Knowledge is protected: what the expert technician knew «by ear» becomes a recorded measurement.
  • Return comes from the first avoided failure: on a critical machine a single unplanned stop can be worth more than the yearly cost of the system.
What it looks like day to day

From the machine fleet to the work order, in three screens.

Interface examples: one useful alarm is worth more than a thousand charts. Click the screenshot to view it full screen.

Nexus Predict · machine fleet
84%

of jobs are planned: emergency work becomes the exception.

9 out of 10

alarms confirmed in the quarter, with only one false alarm.

Priorities

Press PR-02 goes to the top of the list, with the job already proposed for Saturday's planned stop.

Demo interfaces in Italian with sample data: plant, people and values are made up.

The return

What is one avoided failure worth on your critical machine?

Return is calculated machine by machine: compare what an unplanned failure costs with what the same job costs when planned. The difference is the value of one avoided failure. The starting values are the guide's example: a critical machine running two shifts. Replace them with your own.

  1. What a stopped machine costs. Enter the cost of one hour of downtime.
  2. The unplanned failure. How many hours it stays down and what an emergency repair costs.
  3. The same job, planned. How many hours it takes in a planned stop and what it costs.
1Your critical machine
Lost margin on parts not produced, plus overtime to catch up.
2If it fails without warning
On two 8-hour shifts (16 h a day) that is 3 days of lost production.
Emergency technician, urgent shipments, knock-on damage to shaft or gearbox.
3If you act first, in a planned stop
The job is done during an already planned stop, with the spare part ready.
Value of one avoided failure
69,000 €

87,000 € unplanned failure − 18,000 € planned job

Unplanned failure

Downtime · 48 h × 1,500 €/h
72,000 €
Repair and spares
15,000 €
Total
87,000 €

Planned job

Downtime · 8 h × 1,500 €/h
12,000 €
Repair and spares
6,000 €
Total
18,000 €
In a year, with 1 failure avoided69,000 €

To compare with the cost of the system: sensors and installation, connectivity, platform or licences, integration with existing systems, people's time to analyse alarms, and training.

Where to start in your sector

The assets where a stop weighs most, sector by sector.

  • Spindles
  • Axes and ball screws
  • Hydraulic power units
  • Compressors

The indicators to track

Measure them before and after the pilot.

Unplanned stops (hours and number)The main effect of predictive maintenance
MTBF, mean time between failuresAsset reliability
MTTR, mean time to repairFalls with planned jobs and spares ready
Planned jobs over totalHow much work is scheduled rather than urgent
Maintenance cost per asset or per partSpending efficiency
Correct and false alarmsModel quality, shop-floor trust
Spare-parts stock valueEffect on inventory
OEE, availability componentEffect on production
Connections

Predictive maintenance pays off more when it is not an isolated system.

MES, maintenance software, spare-parts store, quality and the safety calendar: each one gives and receives something.

Spare parts ready, work orders in the CMMS and safety deadlines in the same flow.

MES

Production context (order, item, speed, load), running hours, stops and causes, intervention windows.

CMMS

Work orders created from alarms, job history that trains the models.

ERP and stores

Spare-part availability and purchasing, job costs.

Quality

Scrap and process drift as early warning signals.

Safety calendar

Mandatory maintenance, periodic inspections and check registers, which stay on schedule.

Energy monitoring

Abnormal consumption as a symptom of degradation.

Workplace safety

Italian Legislative Decree 81/2008, art. 71

Employers must keep equipment in suitable condition through adequate maintenance and subject it to the required checks and inspections. Predictive maintenance reduces sudden failures, which are also a risk for operators, but it doesn't replace mandatory periodic checks: it must be integrated with the calendar of legally required maintenance and inspections.

Incentives

Hyper-depreciation 2026–2028

For investments from 1 January 2026 to 30 September 2028 the incentive in force in Italy is hyper-depreciation (Law 199/2025): an increase in the cost recognised for income-tax purposes of 180% up to €2.5 million, 100% up to €10 million and 50% up to €20 million. A predictive maintenance system can qualify in two ways:

  • Software (Annex V): it expressly lists predictive maintenance software and algorithms, CMMS systems and software for monitoring machine working conditions.
  • Tangible assets (Annex IV): systems for monitoring and controlling machine working conditions and devices for adding sensors, including on existing plants.
  • Conditions: new assets, interconnected with the production management system, a sworn technical appraisal and notifications to the GSE. How the software is acquired (ownership, licence or subscription) also matters: check case by case with your accountant and appraiser.
How to start

Seven phases, from the most critical machine to the whole fleet.

Criticality analysis

A list of assets with, for each one, the impact of a stop on production, the cost of failures in recent years, redundancies and spare-part lead times. Choose the 3–10 assets where a failure costs most.

Where to start

Failure modes

For each chosen asset: how it usually fails, which signals anticipate the failure, which monitoring technique sees them.

What to look for

Data already available

What PLCs, CNCs and the MES expose; what is missing and needs sensors.

Before buying

Pilot

Sensor installation, a few weeks collecting normal behaviour, first thresholds and alarms.

A few weeks

Response process

Who receives the alarm, how quickly they assess it, how it becomes a work order, how the outcome is recorded.

With maintenance

Measuring results

Stops avoided, false alarms, costs compared with the starting point.

End of pilot

Scaling up

To other assets, other plants and more advanced models.

Step by step

Checklist for choosing a solution

Tick the requirements the solution you are evaluating meets

0 / 8
  • Sensors suited to the environment (temperature, dust, humidity, ATEX zones) and easy to install

  • Reading data already in the machines, as well as new sensors

  • Clear alarms, with the likely cause and urgency

  • Integration with maintenance software and the MES

  • History kept for years and exportable in open formats; data owned by the company

  • Cybersecurity of sensors, gateways and platform

  • A proportionate cost model: per-asset subscription for an SME, a scalable platform for a group

  • Support from expert analysts in the first months, until the team is autonomous

Mistakes to avoid

Monitoring everything at once
Consequence: High costs, too much data, no prioritiesHow to avoid it: Start from the critical assets chosen in the analysis
Buying sensors without knowing which failures to look for
Consequence: Data that anticipates nothingHow to avoid it: Start from the failure modes
Alarms with no response process
Consequence: Ignored alarms, lost trustHow to avoid it: A named owner, response times and defined work orders
Too many false alarms
Consequence: The team stops looking at themHow to avoid it: Thresholds reviewed in the first weeks, production context from the MES
Job outcomes not recorded
Consequence: The model does not learnHow to avoid it: Closing the work order with the real cause of failure
Expecting artificial intelligence from day one
Consequence: Disappointment, a stalled projectHow to avoid it: Thresholds and anomalies first, then predictive models
Data only in the machine builder's platform
Consequence: History not available to the companyHow to avoid it: Access and export guaranteed in the contract
How we work

One useful alarm is worth more than a thousand charts.

Our goal is for every signal collected to reach the right person, with a clear indication, and become a planned job.

People

We start from the experience of maintenance technicians and shop-floor managers: they know how machines break. Together we define who receives the alarms and how to respond.

Systems

We collect data from sensors and machine controllers, link it to the production context in the MES and create work orders in the maintenance software.

Progress

We start with thresholds and anomalies on critical assets, measure stops avoided and false alarms, and introduce predictive models when the history allows it.

  • Criticality analysisAssets and failure modes, to choose where it pays to start.
  • Data collectionReading signals already in PLCs and CNCs, sensors where needed, gateways and a historical archive.
  • Software developmentDashboards, alarm rules, anomaly detection models, asset reports.
  • IntegrationWith MES, maintenance software, ERP and spare-parts store.
  • Legal calendarLinking to mandatory maintenance, which stays on its schedule.
  • Data for hyper-depreciationInterconnection, continuous monitoring and history that help prove the requirements of incentivised assets.
For SMEs and large companies
  • SMEs: a few critical assets, wireless sensors, clear alarms by smartphone or email, help interpreting the first months.
  • Large companies: a platform for asset families across sites, integration with group CMMS and ERP, a data foundation for remaining-life models.
Our commitments
  • Start from the failures to prevent, not the sensors to sell.
  • Data owned by the company, accessible even when the machine builder also monitors it.
  • Transparency on results: we measure correct alarms and false alarms.
  • Predictive maintenance adds to legal checks, it doesn't replace them.
The Emilia-Romagna context

The land of machine builders: those who make them and those who use them.

Between Modena, Reggio Emilia, Parma and Bologna there are districts building machines sold worldwide and companies using them every day. Predictive maintenance matters to both: builders can offer data-driven service, users can cut unplanned downtime.

€22.4bn

Emilia-Romagna machinery and equipment exports in 2024, 26.8% of the regional total 1

€21.7bn

2024 exports of the monitored Emilia-Romagna industrial districts 2

Pilot line

in Bologna, at the BI-REX competence centre: condition monitoring and machine diagnostics training 3

Province of

Modena

  • Machinery: €4.6bn of exports in 2025. 4
  • Sassuolo tile district and Mirandola biomedical cluster (2024 exports +4.6%). 2
Province of

Reggio Emilia

  • Mechatronics district and, with Modena, farm machinery. 2
  • Dairy district: exports +18.9% in 2024. 2
Province of

Parma

  • Food machinery district: exports +14.8% in 2024. 2
  • Food (+2.7%) and dairy (+31%) districts. 2
Province of

Bologna

  • Packaging machinery district. 2
  • BI-REX: predictive maintenance courses with vibration analysis and tests on the pilot line. 3

Where we work. We are based in Modena and work on site with companies in the provinces of Modena, Reggio Emilia, Parma and Bologna. We also work with companies in other provinces and regions, remotely or with agreed visits: write to us from anywhere in Italy.

Glossary

Maintenance terms, explained in one line.

Predictive maintenance
A job planned on the basis of a failure forecast derived from data on the asset's condition.
Condition-based maintenance
A job decided on the basis of a condition check, without forecasting how it will evolve.
Prescriptive maintenance
Predictive maintenance that also suggests the best action and timing.
Critical asset
A machine or component whose failure has the greatest impact on production, costs or safety.
Failure mode
The way a component fails: wear, breakage, leak.
RUL
Remaining Useful Life: the estimated remaining life.
MTBF
Mean time between two failures.
MTTR
Mean time to repair.
CMMS
Maintenance management software: assets, plans, work orders, spare parts.
Edge
Data processing close to the machine, on the shop floor.
Spectral analysis
Breaking vibration down by frequency to recognise the type of defect.
False alarm
A degradation warning that does not correspond to a real problem.
Frequently asked questions

The questions we hear most often.

14 questions

What is predictive maintenance?

It is the strategy that uses data measured on machines and plants to predict when a component will fail and plan the job just before, avoiding both the failure and premature replacements.

What's the difference between preventive and predictive maintenance?

Preventive maintenance acts at fixed intervals of time or hours, even if the component is still good. Predictive maintenance acts on the real condition and the degradation trend.

Is predictive maintenance suitable for an SME?

Yes. With a few wireless sensors on critical machines and a cloud platform, an SME can start at modest cost. Often a single avoided stop on the bottleneck machine pays for the system.

What are the advantages for a large company?

It lets you manage hundreds of assets by risk priority, compare similar plants across sites, integrate alarms, work orders and spare parts, and build reliable remaining-life models thanks to the volume of data.

Do you really need artificial intelligence and large amounts of data?

No. You start with thresholds and anomaly detection, which need a few weeks or months of data. Remaining-life models come when the history of failures and jobs is sufficient.

Which sensors are needed?

It depends on the failures to prevent: accelerometers for vibration, temperature probes, current measurements, pressure sensors, ultrasound. Many useful signals are already available in modern machine controllers.

Can it be applied to old machines?

Yes. External sensors can be fitted to machines without advanced electronics; in fact, the oldest and most critical machines often benefit the most.

How long before the first results?

After installation it takes a few weeks to learn normal behaviour. The first useful anomalies can emerge within the first months.

Does predictive maintenance replace legally required checks?

No. Periodic inspections, fire-safety checks and maintenance required by law keep their deadlines. Predictive maintenance adds to them and reduces failures between one check and the next.

Can a predictive maintenance system qualify for incentives?

Yes, it can qualify for Italy's 2026–2028 hyper-depreciation (Law 199/2025), for investments from 1 January 2026 to 30 September 2028. Annex V on eligible software expressly lists predictive maintenance software, CMMS systems and monitoring of machine working conditions; Annex IV on tangible assets lists machine monitoring systems and devices for adding sensors, including on existing plants. Assets must be new and interconnected with the production management system, with a sworn technical appraisal and notifications to the GSE. How the software is acquired (ownership, licence or subscription) also matters: check case by case with your accountant and appraiser.

How is the return on investment calculated?

For each asset, compare the cost of an unplanned failure (downtime, emergency, collateral damage) with that of a planned job, multiply by the failures avoided in a year, and add the savings on spares and unnecessary maintenance.

Who should handle the alarms?

A named owner: in SMEs the shop-floor manager or the technician, even external; in large companies the maintenance office or a monitoring centre. Without a response process, alarms lose their value.

Why choose Nexus for predictive maintenance?

Because it covers the whole chain: analysis of critical assets, data collection from machines and sensors, analysis software, integration with MES and maintenance, and defining the alarm response process. It starts small and grows on measured results.

Where can you see predictive maintenance in practice in Emilia-Romagna?

In Bologna the BI-REX competence centre has a pilot line where it runs condition monitoring and diagnostics training, for example vibration analysis on bearings and gears. For a project in your company, start from one or two critical machines and the data that already exist.

Let's talk

Do you have a machine that cannot stop?

Let's start there: we analyse how it fails, which data is already available and what is needed to get a warning before the next stop. We reply in person, usually within one working day.

info@nexus-software.it