Predictive maintenance and Industry 4.0 use cases

From fixing failures to predicting them

Every factory, water works and building has machines that wear out: motors, pumps, fans, compressors, conveyors. There are four broad ways to look after them.

Reactive maintenance is cheap until something breaks at the worst time. Preventive maintenance replaces parts on a calendar, which often throws away parts with plenty of life left. Predictive maintenance (PdM) watches each machine's actual condition and plans work when the data shows deterioration. Prescriptive maintenance goes a step further and recommends what to do, for example "reduce speed and replace the bearing at the next planned stop".

The value comes from fewer unplanned stoppages, fewer spare parts held "just in case" and safer work, because repairs happen in planned conditions rather than emergencies. It does not remove the need for skilled technicians; it tells them where to look.

What gets measured

Analytics usually start simple. A threshold on overall vibration catches obvious problems. Trend analysis shows a value creeping up over weeks. Anomaly detection learns what "normal" looks like for one machine and flags unusual patterns. With enough failure history, models estimate remaining useful life (RUL). A rolling statistical check is a good first experiment on your own data:

import statistics

def is_anomaly(history, value, limit=3.0):
    """Flag a reading more than `limit` standard deviations from recent history."""
    if len(history) < 30:
        return False
    mean = statistics.fmean(history)
    sd = statistics.pstdev(history) or 1e-9
    return abs(value - mean) / sd > limit

Other Industry 4.0 use cases

Industry 4.0 describes the joining of physical production with data, connectivity and automation. Predictive maintenance is one part. Others you will meet include:

• Overall equipment effectiveness (OEE) dashboards that combine availability, performance and quality per line.
• Automated quality inspection with cameras and machine-learning models that spot defects.
• Digital twins: live software models of a machine or line used for what-if testing.
• Energy optimisation, shifting loads and finding waste such as air leaks.
• Asset tracking of tools, pallets and mobile robots across a site.

Doing it securely

Predictive maintenance means pulling data out of operational technology (OT) networks, often into cloud analytics. The danger is creating a new route in. Follow these principles, which line up with IEC 62443 and the Purdue model:

1. Prefer separate condition-monitoring sensors or read-only access to PLC data through a historian or gateway in a demilitarised zone (DMZ), never direct cloud access to controllers.
2. Make connections outbound only from the plant. Where risk is high, use a hardware data diode that physically allows one-way flow.
3. Authenticate gateways with certificates and encrypt traffic with TLS.
4. Keep vendor remote access separate, time-limited and logged.
5. Treat models and dashboards as safety-relevant: a poisoned data feed could hide a real fault, so validate inputs and keep engineers in the loop.

Tip: in interviews and exams, contrast preventive (time-based) with predictive (condition-based) maintenance, then explain the security trade-off: more data out of OT means more connections, so design them one-way and through a DMZ.