Skip to content

Services · Digital & automation

Predictive maintenance and machine monitoring

We measure pressure, temperature, vibration and current on your machine, train a model of its normal operation and warn you when it starts to drift — naming the component and leaving time to plan the repair.

What we do

  • Machine review: what it already measures, what is missing
  • Sensor selection and fitting, measurement enclosure
  • Reading controller data without changing its program
  • A model of normal operation for each machine
  • Alarm naming the component, with a recommendation
  • Forecast: a time range until repair is needed
  • Machine screen: condition, trends, repair history
  • Repair after the alarm by the same service team

What predictive maintenance is

A machine rarely fails without warning. Between the first measurable symptom and the stop there are days, weeks or months. Predictive maintenance means measuring that symptom and using the time that is left.

Deviation from normal operation

1 2 3

Running time Window for repair

  1. The model detects a deviation

    Signals leave the band of normal operation for this machine. A single gauge still shows nothing.

  2. Fixed alarm threshold

    This is where an ordinary alarm set to a fixed value would trip — usually shortly before the stop.

  3. Failure

    The machine stops. Without measurement this is the first moment anyone knows about the fault.

Illustrative diagram, not a recording from a specific machine. The length of the window depends on the component and the type of fault.

Example applications

The sensor set and the model depend on the equipment. Below are the three we meet most often in service work.

Hydraulics

Hydraulic power unit and pump

We measure

  • Delivery pressure and its ripple
  • Case drain flow from the pump housing
  • Oil level and temperature in the tank, return temperature
  • Pressure drop across the filter
  • Oil cleanliness class to ISO 4406 and water content
  • Motor current

The model detects

  • Pump wear: falling volumetric efficiency
  • A filter clogging before its indicator trips
  • Cavitation and aeration of the oil
  • Oil ageing and an underperforming cooler
  • Oil loss, meaning a leak, before it shows on the floor

What you gain

You recondition the pump during planned downtime — before debris from its wear spreads through the system and damages valves and cylinders.

Cyclic machine

Hydraulic press and cylinder

We measure

  • Pressure trace of every cycle
  • Pressure build-up time and cycle time
  • Pressure decay while holding
  • Rod position
  • Oil temperature

The model detects

  • Internal leakage: worn piston seals
  • A valve that sticks or fails to close
  • Loss of nitrogen pre-charge in the accumulator
  • A lengthening cycle before the operator notices

What you gain

It ends with a seal change rather than reconditioning the rod. A drop in holding force shows in the data before it shows on the part.

Drive

Motor, gearbox, bearings

We measure

  • Vibration: RMS velocity and spectrum
  • Bearing and housing temperature
  • Motor current on three phases
  • Speed and load

The model detects

  • Damage to bearing races and rolling elements
  • Imbalance and misalignment
  • Looseness and slack fixings
  • Gear tooth wear

What you gain

You order and change the bearing during planned downtime. One that seizes in service often takes the shaft or the housing seat with it.

From sensor to warning

  1. Set-up

    signals
    1

    Measurement

    We use what the machine and its controller already measure. Missing sensors are fitted during planned downtime.

  2. a few weeks of data
    2

    Data collection

    Signals are stored together with the operating state: product, load, ambient temperature.

  3. model of normal operation
    3

    Model training

    The model learns the normal operation of this particular machine. We verify it on data not used for training.

  4. Continuous operation

    alarm with recommendation
    4

    Prediction

    A deviation indicator is calculated continuously. The alarm names the signal, the component and a recommendation for maintenance.

  5. 5

    Machine screen

    Component condition, trends, alarms and repair history in one place — on a computer and on a phone.

Identified faults feed back into model training

What you get out of it

Downtime when you choose

You repair when the part is on the shelf and the crew is on site — not at night, in the middle of an order.

A smaller repair

A fault caught early is seals or a filter. Caught late, it is a rod to recondition, a new pump and a system flush.

Parts without the rush

You order the component the normal way, for a specific fault. No express surcharges and no standing idle waiting for delivery.

Inspections by condition

You replace what has worn, not what the calendar says. Fewer needless replacements and fewer missed faults.

Three ways to maintain a machine

After failureBy the calendarBy condition (predictive)
When you repair When the machine stopsAt a set interval or cycle countWhen the data shows wear progressing
Downtime Unplanned, at the worst momentPlanned, sometimes unnecessaryPlanned, at a time you choose
Parts Ordered as an emergencyReplaced even when still goodOrdered ahead, for a specific fault
What it needs NothingAn inspection scheduleSensors, data and a model
When it makes sense Cheap equipment with a spare on the shelfWear that is predictable over timeA critical machine or one costly to repair

Which models we use

Not one model for everything, but several levels — each needs more data than the one before.

Thresholds and trends — from day one. Limits come from standards and the maker’s documentation: vibration zones to ISO 20816, oil cleanliness class to ISO 4406, permitted operating temperature. Rate of change is added to that — a signal rising week after week matters more than a single high reading.

Anomaly detection — after a few weeks. The model learns from normal operation only, so it needs no failure history. It learns how the signals relate to each other across operating states: how much heat and what pressure go with a given load. That is how it tells oil that is warmer because it is summer and a heavier product is running from oil that is warmer because the pump is losing efficiency. The output is a single deviation indicator plus the list of signals driving it.

Cycle signature — for presses and cyclic machines. Every cycle is compared with a reference: pressure build-up time, peak value, decay while holding, return time. Seal or valve wear shows as a slow drift in those values long before the press stops holding pressure.

Diagnosis and forecast — once a history has built up. After a few identified faults the model starts to tell their types apart, and from the trend of the indicator we estimate how much running time is left before repair. We give a range, not a date.

Models that predict product quality, order lead time or power demand rather than machine condition are covered separately: machine learning in manufacturing.

What the machine screen shows

  • Condition of each component: normal, watch, plan a repair
  • Deviation indicator with its history and the signals driving it
  • The latest cycle against the reference — for presses and cyclic machines
  • Alarm with a recommendation: what to check and by when
  • Cycle and running-hour count since the last repair
  • History of inspections, repairs and replaced parts

How we connect to the machine

Sensors are added alongside the existing controls and brought into a separate measurement enclosure. If the controller already measures what we need, we read the data from it — without changing the machine program. That is why monitoring also works on older machines: a pressure and temperature sensor on a power unit does not need a modern controller.

Slow signals such as temperature are stored every few seconds. Cycle pressure and vibration are sampled far more densely, because wear shows in the shape of the trace. Data is buffered at the machine, so a network outage leaves no gap in the history. Where it is stored is agreed before we start.

Adding a sensor can count as a modification that needs a safety assessment. We do that as part of our machine safety and CE marking work.