Contents
Corrective maintenance has a presentation problem and a measurement problem. The presentation problem is that the methodology literature treats corrective maintenance as the strategy that mature organisations are migrating away from — toward preventive, then predictive, then prescriptive — implying that the share of corrective work should be driven toward zero. The measurement problem is that the share is hard to measure honestly: the reported corrective-versus-preventive split comes from work orders, and work orders do not capture everything that happens on the shop floor.
This article explains what corrective maintenance is, why the reported split and the real one can diverge, when run-to-failure is the cheapest correct strategy, and what separates expensive from cheap corrective work.
What corrective maintenance actually is
Corrective maintenance is the work that happens after a fault, failure, or out-of-spec condition has been detected, with the goal of restoring the asset to its required functional state. It is commonly divided into immediate corrective maintenance (response to a stoppage with safety or production-blocking consequences) and deferred corrective maintenance (the fault is real but the asset can keep running until the next planned window); EN 13306 uses these terms. Run-to-failure is not a sub-type of corrective maintenance but a deliberate maintenance strategy in which an asset is operated until it fails and then repaired or replaced, because the economics of preventive intervention do not justify the cost. Its repair work is corrective, but the decision behind it is strategic, and treating it like a breakdown muddies internal reports.
Around corrective maintenance sits the standard infrastructure: CMMS for work-order routing, MTTR and MTBF as the headline reliability metrics, RCM and FMEA as the analytical frameworks for deciding which assets get which maintenance strategy, and CAPA as the closure discipline that is supposed to convert each significant failure into a learning that prevents recurrence. The infrastructure exists. Whether it is being used to produce honest numbers and honest decisions is a separate question, and the answer is usually less reassuring than the org chart suggests.
The measurement honesty problem
The reported maintenance mix and the actual one can diverge. Three mechanisms explain how.
First, work that should be classified as corrective gets logged as preventive because it happened during a planned-maintenance window. A pump that failed two weeks before its next scheduled service, was kept limping along until the service date, and then got replaced during the window — that is corrective maintenance with deferred timing. It can end up in the CMMS as preventive, because that is the type of the work order it was attached to. The result is that preventive numbers are inflated and corrective numbers deflated.
Second, small unscheduled interventions by operators or front-line technicians — clearing a jam, adjusting a sensor that started drifting, replacing a worn tool earlier than planned — are corrective work in every meaningful sense, but they often never become work orders at all. The operator just fixes the thing and continues. The maintenance log shows none of it. The OEE log shows the downtime, sometimes, depending on how the line is instrumented. The two systems do not reconcile, and the corrective-work fraction in the maintenance system understates reality by whatever the front-line autonomous work amounts to.
Third, there is an organisational incentive to under-report corrective work because the official KPI direction is to reduce it. Departments do not enthusiastically report numbers that make them look like they are heading the wrong way. This is not malicious; it is the same selection bias that affects every metric whose direction-of-travel is itself a reportable target. The fix is not to stop tracking the metric. The fix is to measure the underlying reality through a channel that is not the same channel as the reporting incentive.
When corrective is actually the right strategy
The implicit assumption in the maintenance-maturity-curve literature is that more preventive is always better than more corrective, and predictive is better than both. This is not true at the level of an individual asset. It is true on average across a portfolio, but the average obscures a strategically important truth: for a meaningful subset of assets in any plant, run-to-failure is the genuinely cheapest correct strategy, and forcing them into a preventive regime makes the maintenance budget worse rather than better.
The asset classes where corrective maintenance — specifically, run-to-failure — is the right answer share a small number of characteristics. The asset is not safety-critical, so failure does not endanger people. The asset is not production-critical, either because there is redundancy, because its function is non-blocking, or because the failure can be tolerated until the next scheduled window. The cost of a failure event (the part, the labour, the limited downtime) is materially lower than the cost of running a preventive program over the same time horizon (regular inspection labour, replacement of components that still had useful life, planning overhead). And the failure mode is something the organisation is set up to handle quickly when it does happen — spare parts on hand, a technician who knows the unit, a documented procedure.
Typical examples: small auxiliary motors that drive non-critical conveyance, low-cost sensors whose failure is detected immediately and whose replacement takes minutes, lighting and signalling components, certain classes of pneumatic fittings, some commodity bearings on non-critical rotating equipment. Forcing any of these into a preventive maintenance program produces a calendar of inspections and early replacements that costs more than letting the failure happen and responding to it. The maintenance organisation that has not made this distinction explicitly tends to be running an over-engineered preventive program on assets that did not need one and an under-engineered corrective response on the assets that actually do.
What separates expensive corrective from cheap corrective
For the assets where corrective maintenance is the right strategy — and for the considerable share of corrective work that happens regardless of strategy because plants are not in steady state and reality intervenes — the cost of a corrective event varies enormously between plants in ways that are not about the failure itself. The technical fix is usually similar everywhere. What varies is the surrounding organisational machinery. Factors that make corrective maintenance cheaper include: spare parts physically available at the point of need rather than in a central store far from the line, work-order routing that reaches the right technician promptly rather than at the next shift handover, machine data that lets the technician arrive with a hypothesis about the failure rather than starting diagnosis from zero, and a closure discipline that captures the failure mode in enough structured detail that the next occurrence can be diagnosed faster.
None of those are about the maintenance strategy in the abstract. They are about the operational support layer that determines how long a corrective event takes. In plants where the support layer is weak, the official response is almost always to push harder on preventive maintenance — which is the wrong fix, because the preventive program will not catch the random failures that drive the highest-cost corrective events anyway, and it diverts budget and attention from the support-layer investments that would actually reduce the cost of the corrective work that is going to happen no matter what the strategy says.
Where machine data changes the picture
The single most useful contribution that the data layer makes to corrective maintenance is not predictive analytics on failure modes — that is the headline use case in the literature, and it is genuinely valuable for a subset of high-value assets — but the much more mundane improvement of giving the corrective response itself the context it needs to be fast. When the technician's first action on arrival is "what was the machine doing in the ten minutes before the stop," and that question is answered by pulling up the actual cycle and parameter trace rather than by interviewing the operator, diagnosis can get shorter, which shows up in the MTTR when you compare it before and after the change. This holds whether the underlying maintenance strategy is reactive, preventive, or predictive: better data can make corrective work cheaper independently of strategy.
The second contribution, less universally appreciated, is honest measurement. When the unscheduled-stop log from the line-side data acquisition is cross-referenced against the CMMS work-order log, the gap between official corrective fraction and actual corrective fraction becomes visible — and once it is visible, the organisation can have an honest conversation about whether to reduce the gap (by formalising the work that is happening informally) or to accept it (by acknowledging that some corrective work will always live in the operator's hands and not in the maintenance system). Either answer is workable. Pretending the gap does not exist is the only answer that is not.
In SYMESTIC, the functions that touch corrective maintenance most directly are Alarms (alarm and fault messages by push or e-mail), downtime reasons with MTBF and MTTR, and Process Data (the parameter trace from the minutes before a stop). SYMESTIC is not a CMMS: work orders stay in the maintenance software, and the data is handed over to it via REST API. The platform does not decide on its own whether a given asset belongs on corrective, preventive, or predictive maintenance, because that decision has to be made by the people who know the asset and the cost structure around it. What it provides is measured data for that conversation: stop reasons, MTBF and MTTR. Once those numbers come from measurement instead of estimates, the strategy discussion has a factual basis.
Potential analysis30 minutes · free
Put a euro value on your breakdowns
In a free 30-minute call we take your unplanned stops and repair times and calculate their annual cost. No software demo.
In use in over 20 countries
Loss statementYour plant
More than 30 years in industrial automation and machine connectivity, from Simatic S5 to OPC UA. Dipl.-Ing. in communications engineering.
- Machine Data Collection (MDC)
- Operational Data Collection
- Machine Connectivity
- PLC Programming (S5 / S7 / TIA)
Updated
All articles by Martin Brandel
