Preventive Maintenance: Definition, Types & KPIs 2026

What is preventive maintenance?

Preventive maintenance (PM) is a scheduled maintenance strategy that performs inspections, servicing and component replacements at fixed intervals — defined by calendar time, operating hours, cycles or produced units — in order to reduce the probability of unplanned equipment failure. It is proactive by design: work is triggered by a plan, not by a breakdown. In German the term is vorbeugende Instandhaltung or präventive Wartung.

The underlying economic logic is simple. Unplanned downtime is always more expensive than planned downtime — typically 3 to 10 times more per hour, because it happens at the worst possible moment, blocks the bottleneck, pulls operators off other work and often propagates through quality and delivery. Preventive maintenance trades a smaller, predictable cost (the planned stop) against a larger, unpredictable one (the breakdown). The strategy works when the planned cost is genuinely smaller than the probability-weighted avoided cost. When it isn't, PM becomes over-maintenance — and that failure mode is as common as under-maintenance.

Where does PM sit in the maintenance strategy ladder?

Maintenance strategies sit on a ladder of increasing data intensity. Preventive maintenance is the second rung. Understanding the full ladder makes it clear what PM solves and what it does not.

The industrial reality in 2026: most manufacturing plants run a mixed portfolio. Reactive on consumables and non-critical parts. Preventive on the majority of production assets. Condition-based and predictive on the top 5–15% of critical and capital-intensive equipment. The mix is an economic decision, not a technology decision — and getting the mix right is where most maintenance programs either pay back or quietly bleed money.

What are the main types of preventive maintenance?

Inside the preventive category, two sub-strategies dominate. The choice between them drives everything downstream — scheduling, spare parts, labor planning and the quality of the data the maintenance team actually needs.

• Time-based PM. Work is triggered by calendar time: every 30 days, every quarter, every year. Simple to schedule, simple to audit, robust when an asset's duty cycle is stable. Weak when utilization varies — a machine running 20 hours per day and one running 6 hours per day receive the same service on the same date, which is either wasted effort or insufficient cover.
• Usage-based PM. Work is triggered by a counter: operating hours, cycles, produced units, kilometers, tonnes. Matches wear to actual use. Requires a reliable counter signal from the asset — which is where MES-level machine data capture becomes decisive, because "reliable counter" means captured automatically, not self-reported by operators on a paper log.

Usage-based PM is the honest default for production equipment. Time-based PM is the fallback when the counter is not available or not trustworthy. In brownfield plants with a mix of modern and legacy assets, both coexist — and the gradual migration from time-based to usage-based as machine connectivity improves is one of the most reliable sources of maintenance savings in the first year after MES introduction.

What activities does preventive maintenance actually cover?

The scope of PM is broader than "replacing parts on a schedule," even though that is the most visible activity. A well-structured PM program covers four activity types, and weakness in any one of them shows up as unplanned downtime that the planned work was supposed to prevent.

• Inspection. Visual checks, measurement of wear indicators, functional tests. Low-cost, high-frequency. The early-warning layer.
• Servicing. Lubrication, cleaning, calibration, tightening, parameter checks. Medium-frequency. The condition-stabilizing layer.
• Component replacement. Scheduled swap of known wear parts — filters, belts, seals, tooling — before failure probability crosses an economic threshold. The risk-mitigation layer.
• Overhaul. Major scheduled refurbishment of assemblies or whole machines. Low-frequency, high-cost, planned during production shutdowns.

Most PM programs concentrate budget and attention on component replacement and overhaul — the visible, expensive end of the scope. The highest return usually comes from disciplined inspection and servicing at the cheap, high-frequency end, because that is where small deviations are caught before they compound into failures.

What metrics measure PM performance?

Four KPIs cover the legitimate uses of "maintenance performance" in industrial practice. Together they separate a PM program that is actually preventing failures from one that is producing paperwork.

• MTBF (Mean Time Between Failures) — average operating time between unplanned stoppages on a given asset. The direct measure of whether PM is working. Rising MTBF means prevention is succeeding. Flat or falling MTBF after PM hours go up is a red flag.
• MTTR (Mean Time To Repair) — average duration of a repair once a failure has occurred. Not a PM metric directly, but essential context: a PM program that cuts failures in half while MTTR doubles has not improved availability.
• Planned Maintenance Percentage (PMP) — planned maintenance hours ÷ total maintenance hours. A mature program typically runs 70–85% PMP. Below 50% indicates a reactive culture. Above 90% often indicates over-maintenance, not excellence.
• PM Compliance — completed PM tasks ÷ scheduled PM tasks, measured in a defined compliance window (often ±10% of the due date). Below 90% indicates the schedule is aspirational rather than executed.

The cross-check on the whole set is availability — specifically the Availability factor of OEE. A PM program that does not move availability over two to four quarters is not earning its budget, regardless of how good the internal metrics look.

When does preventive maintenance become over-maintenance?

Preventive maintenance is the default strategy for good reasons, but it has a specific failure mode that rarely gets discussed honestly: over-maintenance. Swapping components that still had useful life. Lubricating a bearing that was already properly lubricated. Calibrating instruments that were within tolerance. Each individual over-service costs a little. Aggregated across a plant, the cost is substantial — and the induced failures (every intervention introduces some probability of human error) are often invisible in the PM budget because they show up in the "unplanned" column.

Four warning signs of over-maintenance appear consistently in industrial practice. PM Compliance above 95% combined with flat MTBF — the work is getting done but nothing is improving. High rates of post-PM failures (the "infant mortality" hump that follows aggressive component replacement). Spare parts consumption rising faster than production volume. Maintenance labor cost per produced unit rising while availability is stable. Any two of these together justify a hard look at the PM intervals — and usually at the 20% of tasks that account for 80% of the hours.

How do MES data change the quality of PM decisions?

Preventive maintenance without reliable runtime data is an educated guess about average wear. Preventive maintenance with MES-sourced data is an informed decision about actual wear on each specific asset. The difference matters for three decisions that together determine whether PM pays back.

First, the interval itself. Usage-based intervals require trustworthy counters — captured automatically from the PLC where possible, from digital I/O gateways on brownfield assets where the controller cannot be touched. Manual counter readings drift, round, and miss micro-stops, which biases intervals upward (too conservative, too much cost) or downward (too aggressive, too much risk). Automatic capture closes the gap. Second, the trigger quality. Downtime reasons captured through the MES alarm layer build the failure history that reveals which PM tasks are actually preventing failures and which are hygiene theater. Third, the cross-domain view. A failure that happens at a specific cycle count on a specific tool after a specific changeover type is only visible when production, quality and maintenance data share a single context — which is exactly what an MES-centric architecture provides.

Where does preventive maintenance fit in the SYMESTIC platform?

In the SYMESTIC deployment pattern, the platform is not a CMMS — it is the data layer that makes CMMS-driven PM decisions honest. Runtime counters, cycle counts, stop reasons and process parameters are captured automatically via OPC UA on modern controllers and via digital I/O gateways on brownfield assets. The alarms module and process data module together supply the runtime signals and failure history that usage-based PM and condition-based maintenance rely on. Maintenance execution itself typically runs in SAP PM, IBM Maximo, Ultimo or a similar CMMS, connected via REST API. Across the installed base of 15,000+ connected machines in 18 countries, the recurring observation is consistent: the dominant constraint on PM quality is not the maintenance software itself but the reliability of the runtime data feeding it, which is why the data-capture layer is the decisive architectural choice. For authoritative reading, see the ISO 22400 / IEC 62264 standard on manufacturing KPIs and the VDI 2895 guideline on the organization of maintenance.

FAQ

What is the difference between preventive and predictive maintenance?
Preventive maintenance is triggered by a fixed schedule — calendar time or usage counter. Predictive maintenance is triggered by a model-forecasted failure, based on real-time condition data and historical patterns. Predictive is more precise when the data and the models exist; preventive is more robust when they don't. In practice, most plants run both, applied to different asset classes based on criticality and data availability.

Is time-based or usage-based PM better?
Usage-based is the default when reliable counters exist, because it matches wear to actual use. Time-based is the fallback when counters are absent or untrustworthy. The migration from time-based to usage-based PM is one of the most reliable maintenance savings levers after MES introduction, because automatic runtime capture makes usage-based intervals legitimate for the first time across the whole plant.

What is a good PM Compliance target?
85–95% measured in a defined window around the due date (often ±10%). Compliance below 85% means the schedule is not being executed; above 95% combined with flat MTBF means the schedule is too aggressive and work is being done that isn't preventing anything. The number on its own is meaningless — it has to be read alongside MTBF and availability.

How much of total maintenance should be planned vs. unplanned?
A mature plant typically runs 70–85% planned maintenance. Below 50% signals a reactive culture where breakdowns dominate the day. Above 90% often signals over-maintenance — the ratio looks clean because planned hours have been inflated, not because unplanned hours have fallen. The healthy trajectory is rising PMP combined with falling absolute unplanned hours, not rising PMP by itself.

Does preventive maintenance require a CMMS?
A small plant can run PM from a spreadsheet. A mid-sized or multi-site operation cannot — the combinatorics of assets, tasks, intervals, spare parts and technicians exceed what manual tracking can handle reliably. A CMMS (SAP PM, IBM Maximo, Ultimo and similar) is the standard tool. An MES is not a CMMS, but a connected MES is the data source that makes the CMMS-driven PM schedule trustworthy — usage-based intervals need automatic counters, not hand-written logs.

When does preventive maintenance stop making economic sense?
When the planned cost of a given PM task is larger than its probability-weighted avoided cost, or when usage-based triggers would deliver the same risk reduction at materially lower cost. The review should be periodic — typically annual — and should focus on the 20% of tasks that account for 80% of maintenance hours. Tasks that have not prevented a single failure in two years of data are candidates for interval extension or removal; tasks that coincide with repeated post-service failures are candidates for redesign, not more frequency.

How do I build the data foundation for better PM?
Three layers, in order. First, reliable runtime and cycle counters on every asset that matters — OPC UA where the controller supports it, digital I/O gateways for legacy equipment. Second, structured downtime and alarm capture so the failure history becomes analyzable, not anecdotal. Third, integration with the CMMS so PM triggers, execution records and failure data live in one context. Without layer one, the rest is expensive guessing.

Related: MTBF · MTTR · Availability · OEE · Predictive Maintenance · Condition Monitoring · Manufacturing Operations Management · CMMS · Machine Data Acquisition · Alarms.

About the author

Mark Kobbert

CTO of SYMESTIC GmbH. Responsible for the cloud-native MES architecture since 2014 — Microsoft Azure, microservices, API-first. Built the IoT-gateway and OPC UA connectivity layer that powers 15,000+ connected machines in 18 countries with 99.9% availability. Focus areas: industrial connectivity, brownfield machine integration, edge computing, real-time data processing and IT/OT convergence. B.Sc. Wirtschaftsinformatik, SRH Hochschule Heidelberg. · LinkedIn