Condition monitoring vs predictive maintenance: what the difference is and what a prognosis can deliver

Few pairs of terms in maintenance are confused as often as condition monitoring and predictive maintenance. Many products are called "predictive" but measure a value and compare it with a limit. Others promise a remaining useful life down to the day, without saying what that number is based on. For a purchasing decision this is a problem, because the terms stand for different capabilities. This article sets out the four maintenance strategies, explains the processing stages according to ISO 13374 and describes what a prognosis on a CNC spindle can credibly deliver today.

Four maintenance strategies at a glance

Maintenance terminology is defined in the European standard EN 13306. For practical purposes, a division into four strategies is enough. They differ in what triggers an action.

Reactive maintenance is not automatically wrong. For an inexpensive, easily accessible component with no secondary damage, it can be the most economical choice. That is rarely true for a main spindle, because bearing damage in its final stage can damage the shaft, the housing or the workpiece. What downtime costs in your own plant is shown in the article What does an hour of machine downtime cost.

Preventive maintenance works with fixed intervals. It can be planned, but it is based on averages. This is already evident in bearing selection: the basic rating life L10 according to ISO 281 is a statistical statement about many identical bearings. It gives the running time that 90 per cent of these bearings will reach. It does not tell you when the one bearing in your spindle will fail.

Condition-based maintenance triggers actions when the measured condition requires it. The tool for this is condition monitoring.

Predictive maintenance goes one step further. It is condition-based maintenance in which the future progression is also estimated. Predictive maintenance therefore requires condition monitoring. Without reliable condition detection there is no reliable prognosis.

The processing stages according to ISO 13374

ISO 13374 describes how data is processed in condition monitoring. It divides the path from sensor signal to recommended action into stages that build on one another. This helps when comparing products: you can quickly see at which stage a system stops.

1. Acquire data: sensor and control signals are recorded; on spindles, mainly the vibration on the spindle housing and the speed from the CNC control.
2. Calculate features: meaningful quantities are derived from the raw signal, such as RMS, crest factor, kurtosis and the amplitudes at the bearing frequencies in the envelope spectrum.
3. Detect condition: the features are compared with a reference, either a limit or a learned baseline. The result is a statement such as "noticeable".
4. Make a diagnosis: the system assigns the deviation to a cause, for example damage to the outer race.
5. Make a prognosis: the future progression is estimated, ideally with a remaining useful life.
6. Give a recommendation: condition, diagnosis and prognosis lead to an action with an urgency.

A system that only compares an overall level with a fixed limit covers stages 1 to 3. A diagnosis requires frequency-based methods such as envelope analysis for rolling bearings. How condition detection in stage 3 works properly is described in the article fixed thresholds, 3 sigma or a learned baseline.

Why a remaining useful life to the day is rarely credible

The idea is tempting: the system reports that the spindle bearing will last another 23 days, and the maintenance team schedules the replacement for day 20. On a CNC spindle, such a number is in most cases not reliable. There are solid reasons for this.

• The progression depends on the future: how quickly bearing damage grows depends on load, speed, lubrication and temperature. These change with every job. A roughing programme with high load stresses the bearing differently from a week of finishing.
• The progression is not uniform: bearing damage passes through phases at different speeds. A line in the envelope spectrum can remain stable for a long time and then grow rapidly.
• Run-to-failure data is missing: a reliable prognosis model would require many fully recorded damage progressions of exactly this spindle and bearing type under similar conditions. A single plant has only a few of them.
• Interventions change the progression: if the bearing is relubricated, the speed reduced or the spindle loaded less, the old projection no longer applies.

A remaining useful life given as a single date is therefore usually false precision. At most, a range with a stated uncertainty would be credible, and even that only after an observation period on the specific machine. Anyone who gives you a fixed number of days without that basis should be able to explain where it comes from.

What we deliver instead: condition, trend, explanation

We deliberately do not give a remaining useful life in days. For every monitored spindle, our system delivers three things that can be checked:

1. The condition: one of four levels, measured against the learned baseline, i.e. the normal state of this spindle, separately for each measuring point and speed range. An alert is only raised once the deviation persists over several measurements.
2. The trend: how the features develop over days and weeks. Whether a deviation remains stable or grows tells the maintenance team more than any projection.
3. The explanation: which feature has changed and by how much, which damage pattern is suspected, such as outer race, inner race, rolling element, cage or unbalance, and what to inspect.

The urgency follows from the level. "Noticeable" means keep an eye on it; no immediate action is needed. "Warning" means schedule an inspection at the next opportunity. "Alarm" means inspect promptly, because the risk of failure is increased. This is less spectacular than a countdown. But it is transparent, and the maintenance engineer keeps the decision on when to intervene. They know the order book, the spare parts situation and the next free maintenance slot better than any model.

Our offering therefore sits clearly on stages 1 to 4 of ISO 13374, supplemented by trend and recommendation. What the system monitors today and what it does not is set out on the page Spindle monitoring for CNC machining centres.

Which strategy suits which machine

Not every machine needs condition monitoring. The choice depends on three questions: what does an unplanned failure cost, how long does it take to obtain a replacement, and does the damage announce itself measurably?

• Reactive is enough for inexpensive parts with no secondary damage that can be replaced quickly.
• Time-based suits wear parts with a well-known service life and inspections required by law or by the manufacturer.
• Condition-based pays off where a failure is expensive, replacement takes time and the damage shows itself in advance. Main spindles of machining centres meet all three conditions: rolling bearing damage announces itself in the vibration signature.

In practice, almost every plant mixes these strategies, often on the same machine. On a machining centre, the coolant is changed according to schedule, a defective limit switch is replaced reactively and the main spindle is monitored on a condition basis. Condition monitoring does not replace the maintenance work prescribed by the machine manufacturer either. It supplements it where fixed intervals come too early or too late.

It is also important to look at the organisation. An alert is only useful if someone reads it, schedules the inspection and reports back on the result. Before you introduce a system, it should therefore be clear who receives alerts, who decides on actions and how findings are documented. This feedback improves the assessment over time, because the thresholds can be calibrated against experience with your machines. A system without this feedback loop stays where it was on day one.

You can make your own estimate for your plant with our calculator. A broader overview of spindle monitoring is given in the spindle monitoring guide.

Frequently asked questions