What Is Predictive Maintenance?

Q: What Is Predictive Maintenance?

Predictive maintenance (PdM) is a maintenance strategy that schedules work from the measured condition of a machine and a forecast of when it will fail. Reactive maintenance fixes machines after they fail; preventive maintenance services them on a calendar whether they need it or not. Predictive maintenance watches the machine itself, spots the early signs of wear in its data and schedules work only when the condition calls for it. The idea rests on a simple observation: most mechanical faults develop gradually. A bearing starts to wear, its vibration changes, its temperature creeps up, and only weeks later does it seize. Maintenance engineers draw this as the P-F curve, the interval between the point a fault becomes detectable (P) and the point of functional failure (F). Predictive maintenance aims to detect the fault early in that interval and plan the repair before the end of it.

Q: Who invented predictive maintenance?

No single inventor. The idea is usually traced to the 1940s, when C. H. Waddington’s wartime analysis of RAF aircraft showed that scheduled servicing could itself cause failures. Vibration monitoring of rotating machines followed in industry from the 1950s and 1960s.

Q: What is the difference between preventive and predictive maintenance?

Timing. Preventive maintenance follows a schedule, such as every 2,000 hours. Predictive maintenance follows the measured condition of the machine and acts only when the data shows a fault developing.

Q: How does predictive maintenance use AI?

To learn what normal looks like. Machine learning models trained on sensor history flag unusual patterns and estimate remaining useful life, catching faults that a fixed alarm limit would miss.

Q: What is predictive maintenance also called?

PdM, or sometimes condition-based maintenance. Strictly, condition-based maintenance acts on current readings, while predictive maintenance forecasts future failure, but the terms are often used together.

Q: How do you start with predictive maintenance?

With one critical asset. Pick a machine whose failure is costly, fit sensors for its main failure modes, collect a baseline and add analytics once you have history. Then extend to similar machines.

How predictive maintenance works

Predictive maintenance (PdM) is a maintenance strategy that schedules work from the measured condition of a machine and a forecast of when it will fail. Reactive maintenance fixes machines after they fail; preventive maintenance services them on a calendar whether they need it or not. Predictive maintenance watches the machine itself, spots the early signs of wear in its data and schedules work only when the condition calls for it.

The idea rests on a simple observation: most mechanical faults develop gradually. A bearing starts to wear, its vibration changes, its temperature creeps up, and only weeks later does it seize. Maintenance engineers draw this as the P-F curve, the interval between the point a fault becomes detectable (P) and the point of functional failure (F). Predictive maintenance aims to detect the fault early in that interval and plan the repair before the end of it.

The four steps

Every predictive maintenance system, from a single pump to a whole plant, follows the same loop.

//
Sense. Sensors measure the machine’s condition continuously: accelerometers for vibration, thermocouples or RTDs for temperature, current transformers for motor load.
//
Collect. An IIoT gateway gathers the data from the sensors, I/O modules and PLCs, often over Modbus or OPC UA, and timestamps it.
//
Analyze. Analytics or AI models compare it with normal behavior to spot developing faults and estimate the remaining useful life.
//
Act. Maintenance is scheduled before the failure, at a convenient time, with the right parts already on site.

Predictive vs preventive and reactive maintenance

Predictive maintenance is one of four broad approaches, and most sites use a mix. Condition-based maintenance sits closest to it: both act on measured condition, but condition-based maintenance reacts when a reading crosses a limit, while predictive maintenance forecasts when it will.

Approach	When work happens	Data needed	Main drawback	Best suited to
Reactive	After a failure	None	Unplanned downtime and secondary damage	Cheap, non-critical items
Preventive	On a fixed schedule or hours run	Run hours, calendar	Replaces parts with life left	Parts with predictable wear
Condition-based	When a reading passes a limit	Live sensor readings	Alarms can come late	Clear single indicators
Predictive	Before a forecast failure	Sensor history plus models	Set-up effort and data quality	Critical, costly assets

The choice is asset by asset. A light fitting is run to failure; a main drive motor or a compressor on a production line is a natural candidate for predictive maintenance.

FAQ · 01

What is condition monitoring?

Condition monitoring is measuring a machine’s health continuously, using parameters such as vibration, temperature, oil quality, noise and electrical current, to detect changes that point to a developing fault. It is the data foundation that predictive maintenance is built on.

Read the full FAQ —

Techniques used to predict failures

Each technique suits certain faults. Large programs combine several, guided by standards such as ISO 17359 for condition monitoring and the ISO 20816 series for vibration.

//
Vibration analysis. The most common technique for rotating machines. Changes in the vibration spectrum reveal imbalance, misalignment, looseness and bearing wear.
//
Thermography. Infrared cameras find hot spots in electrical panels, motors and bearings before they fail.
//
Motor current analysis. The current drawn by an electric motor carries signs of rotor bar, bearing and load faults, measured without touching the machine.
//
Oil analysis. Wear metals, water and contamination in lubricant show internal wear in gearboxes and engines.
//
Ultrasound. High-frequency sound detects leaks, early bearing faults and electrical discharge.

The hardware behind predictive maintenance

The software gets most of the attention, but the data has to be measured and moved reliably first. Sensors on the machine feed analog I/O modules or connect to the PLC, which turns 4–20 mA and voltage signals into numbers. The gateway at the machine is what makes that data available: our IoT edge gateway case study shows fanless gateways relaying machine data to a predictive maintenance platform, and our industrial IoT gateways are built for the factory floor.

Vibration data is the heaviest load. An accelerometer sampled at tens of kilohertz produces far more data than a network should carry, so the gateway or an edge computer usually processes it on site, sending features such as RMS levels and spectrum peaks to the analytics platform. Where AI models are involved, that step increasingly runs on an edge AI computer beside the line, the pattern covered in our guide to AI in manufacturing.

Benefits and limits

Done well, predictive maintenance cuts unplanned downtime, avoids secondary damage when a failing part takes others with it, and stretches the life of parts that a fixed schedule would have replaced early. Spare parts and labor can be planned rather than called in at night.

It is not free. Sensors, connectivity and analytics cost money and effort, and the models need months of good data before their forecasts can be trusted. A model trained on one machine may not transfer to another of a different age or duty. Some teams pair it with a digital twin of the asset, which lets them test what a reading means before acting on it. Most programs start small and earn their place asset by asset.

Quick answers on predictive maintenance

Predictive maintenance: common questions

Who invented predictive maintenance?

No single inventor. The idea is usually traced to the 1940s, when C. H. Waddington’s wartime analysis of RAF aircraft showed that scheduled servicing could itself cause failures. Vibration monitoring of rotating machines followed in industry from the 1950s and 1960s.

What is the difference between preventive and predictive maintenance?

Timing. Preventive maintenance follows a schedule, such as every 2,000 hours. Predictive maintenance follows the measured condition of the machine and acts only when the data shows a fault developing.

How does predictive maintenance use AI?

To learn what normal looks like. Machine learning models trained on sensor history flag unusual patterns and estimate remaining useful life, catching faults that a fixed alarm limit would miss.

What is predictive maintenance also called?

PdM, or sometimes condition-based maintenance. Strictly, condition-based maintenance acts on current readings, while predictive maintenance forecasts future failure, but the terms are often used together.

How do you start with predictive maintenance?

With one critical asset. Pick a machine whose failure is costly, fit sensors for its main failure modes, collect a baseline and add analytics once you have history. Then extend to similar machines.

We supply the gateways, I/O modules and edge AI computers that collect and process machine data, configured as standard with the protocols, storage and operating system the project needs. Tell us your machines, sensors and platform, and we will specify the hardware to match.