Preventive vs Predictive Maintenance: Costs, Data and Best Fit

Introduction

Preventive maintenance asks, “What work should we perform at a planned interval to manage failure risk?” Predictive maintenance asks, “What evidence suggests this asset is moving toward a failure, and when should we act?”

The second question sounds more advanced, but advanced is not the same as appropriate. Some failure modes are age-related and respond well to scheduled replacement. Some can be detected through condition signals. Some are random and cannot be predicted economically. Some assets are cheap and noncritical enough that run-to-failure is rational. A good strategy uses different policies for different risks.

This article provides a selection framework rather than promising that sensors and AI will eliminate downtime. Logic Unit should add approved Titan MMS capabilities and any first-hand examples only after maintenance and product experts review them.

Table of Contents

1. Definitions
2. How the strategies differ
3. Select by failure mode and consequence
4. Data and technology requirements
5. Cost and business case
6. Implementation roadmap
7. Hybrid maintenance strategy
8. Common mistakes
9. FAQs

Preventive, Predictive and Condition-Based Maintenance

Preventive maintenance

Preventive maintenance is planned work performed at defined intervals, commonly based on calendar time, usage or meter readings. Examples include inspection every month, lubrication after a defined number of operating hours, or replacement after a specified cycle count.

The interval may come from manufacturer guidance, regulation, engineering judgment or operating history. The goal is to reduce the likelihood or consequence of failure through planned intervention.

Preventive maintenance is not automatically good. Work performed too frequently creates cost, unnecessary downtime and the possibility of maintenance-induced failure. Work performed too late may not control risk.

Condition-based maintenance

Condition-based maintenance initiates action when an observed condition crosses a defined rule or indicates degradation. A technician may inspect belt wear, measure vibration, review lubricant condition or observe a temperature trend. The trigger can be manual or automated.

Predictive maintenance

Predictive maintenance uses condition and operating data to estimate degradation or failure risk and recommend an intervention window. Techniques range from engineering thresholds and trend models to statistical or machine-learning models. The value is not the model itself; it is giving the organization enough reliable warning to plan an effective action.

Run-to-failure and corrective maintenance

For low-consequence, inexpensive, redundant and easily replaced items, planned run-to-failure can be sensible. Corrective maintenance is the work required after a defect or failure is found. A mature strategy does not pretend every asset deserves predictive monitoring.

How Preventive and Predictive Maintenance Differ

The categories can overlap. A scheduled monthly vibration route is preventive in timing but condition-based in the maintenance decision. A model may recommend intervention, while the CMMS schedules the work and preserves history.

Select the Policy by Failure Mode

The unit of analysis should be the failure mode, not the machine label. One pump can have different policies for bearing degradation, seal leakage, lubrication, electrical supply and accidental damage.

Step 1: Describe the function

State what the asset must do, at what performance and under what conditions. “Pump water” is incomplete if the required flow, pressure, availability and environment matter.

Step 2: Define functional failure

Describe how the asset fails to meet the requirement: stops completely, produces insufficient flow, leaks, consumes abnormal energy, creates a safety risk or damages quality.

Step 3: Identify credible failure modes

Use engineering knowledge, manuals, work history, inspections and operator experience. Avoid a generic list copied from another plant.

Step 4: Assess consequence and criticality

Consider safety, environment, quality, production/service, compliance, repair cost, redundancy and detectability. Criticality should guide effort, not merely decorate the asset record.

Step 5: Ask whether the failure is age-related

If the probability increases predictably with age or usage, a preventive replacement/overhaul interval may be effective—provided the intervention costs less than the managed risk and does not introduce unacceptable failure.

Step 6: Ask whether degradation can be detected

There must be a measurable change before functional failure and enough P-F interval—the time between potential failure detection and functional failure—to confirm, plan, obtain parts, schedule access and execute work.

Step 7: Compare viable policies

• Scheduled restoration/replacement.
• Scheduled inspection or condition monitoring.
• Predictive model.
• Failure-finding task for hidden protection.
• Redesign or operating change.
• Run-to-failure with prepared response.

Select the policy based on risk reduction, practicality and lifecycle economics.

When Preventive Maintenance Is a Strong Fit

Preventive work tends to fit when:

• Regulation, safety policy or warranty requires an interval.
• Consumables or wear items have reasonably consistent useful life.
• Usage is measurable and correlates with deterioration.
• Inspection itself controls risk.
• Failure history is insufficient for a predictive model, but engineering guidance is credible.
• The cost of planned access is low relative to failure consequence.
• The organization needs a disciplined maintenance baseline before advanced analytics.

Preventive maintenance also builds data. Consistent asset identity, completion, findings, labor, parts and failure information can make later optimization possible.

Review intervals. If technicians repeatedly find no degradation, an interval may be too short or the task ineffective. If failure occurs between services, the interval, procedure, operating context or failure-mode assumption may be wrong.

When Predictive Maintenance Is a Strong Fit

Predictive maintenance is promising when:

• The asset/failure consequence justifies monitoring and response.
• A degradation signal exists before functional failure.
• The warning horizon is operationally useful.
• Sensors or inspections measure the relevant condition reliably.
• Operating context—load, speed, product, environment—is available.
• The organization can verify alerts and act within the window.
• Sufficient labeled history, engineering thresholds or comparable evidence exists.
• False-positive and false-negative costs are understood.

Common signals include vibration, temperature, current, acoustic emission, pressure, flow, oil debris and process performance. A signal is not automatically a diagnosis. Experts must distinguish asset degradation from normal operating variation, sensor error and process change.

Data and Technology Requirements

Asset identity

Condition data must map to the correct asset and component. If sensor tags, CMMS assets and ERP equipment IDs conflict, analysis will be unreliable.

Time and operating context

Synchronize timestamps and capture operating state. Comparing a machine at full load with standby behavior can create false alerts.

Condition-data quality

Monitor missing data, calibration, sensor placement, sampling rate, units, drift and connectivity. An analytics model cannot compensate for uncontrolled measurement indefinitely.

Maintenance and failure labels

Work orders should record the observed problem, cause, action and affected component with enough consistency to evaluate alerts. “Fixed machine” is not a useful label.

Integration to action

A predictive alert should have an operational path:

1. Signal/model identifies an exception.
2. Qualified person reviews evidence.
3. Decision creates an inspection or work order.
4. Planner prioritizes and prepares work.
5. Technician records findings/action.
6. Outcome feeds threshold/model evaluation.

Without this loop, predictive maintenance becomes another dashboard nobody owns.

Security and IT/OT boundaries

Industrial connections must follow the organization’s approved architecture and security controls. Define network zones, identity, least privilege, remote access, patching, logging, vendor support and incident ownership with qualified IT/OT security professionals.

Build the Business Case

Start with one failure mode and asset population.

Current-state cost

Estimate:

• Failure events and confidence in the count.
• Production/service loss attributable to the failure.
• Repair labor, contractor, parts and collateral damage.
• Quality, safety or compliance consequence where quantifiable and appropriate.
• Current inspection/preventive effort.
• Planning and spares constraints.

Future-state cost

Include sensors, installation, connectivity, platform/analytics, integration, model/threshold development, expert analysis, training, monitoring, maintenance of the monitoring system and planned intervention.

Addressable benefit

Do not apply a claimed percentage to all downtime. Estimate how many failures have a detectable warning, how often the system could identify them, how many alerts would be acted on, and what consequence could be avoided or reduced.

Use scenarios:

Expected value = events × detection probability × action probability × avoided consequence − total program cost

This simplified model needs sensitivity analysis. If the case only works with perfect detection and action, it is not ready.

Implementation Roadmap

1. Establish maintenance discipline

Clean asset identity, work-order completion, failure coding, PM execution and ownership. A CMMS such as Titan MMS may support this foundation, but the product team must verify relevant capabilities.

2. Select a bounded use case

Choose a critical failure mode with evidence, accessible data and an achievable response. Avoid “predict all failures.”

3. Baseline and instrument

Document current failure and maintenance performance. Validate sensors, sampling and contextual data. Preserve raw evidence needed to investigate alerts.

4. Start with interpretable rules where possible

Engineering thresholds and trends can be effective and easier to validate than complex models. Increase complexity only when it improves the decision materially.

5. Run in shadow mode

Generate alerts without automatically changing operations. Compare them with inspections and failures. Measure false positives, missed events, warning time and analyst workload.

6. Design human review and work integration

Define who reviews, response time, evidence, escalation and work-order generation. Avoid autonomous high-consequence actions without appropriate controls.

7. Evaluate and scale

Scale when the use case produces reliable, actionable warnings and a sustainable economic result. Monitor changes in equipment, sensors, operating context and model behavior.

A Hybrid Strategy Is Usually the Real Answer

A site can use:

• Run-to-failure for low-consequence consumables.
• Scheduled lubrication and statutory inspections.
• Condition routes for important rotating equipment.
• Predictive analysis for a small number of high-value failure modes.
• Redesign for recurring failures no maintenance interval can control.

The CMMS remains the execution and history layer. Condition tools and analytics supply evidence. ERP supports financial/procurement processes. A data platform may combine information for broader analysis. Define ownership and do not duplicate uncontrolled asset masters.

Common Mistakes

• Buying sensors before identifying failure modes and decisions.
• Calling threshold alarms “AI” to increase perceived sophistication.
• Predicting a signal that gives too little time to act.
• Ignoring false-alert workload.
• Training on unrepresentative operating conditions.
• Using poor work-order labels as truth.
• Automating work orders without review or priority governance.
• Measuring model accuracy without avoided consequence or action rate.
• Keeping ineffective PM after predictive monitoring is proven, creating double work.
• Eliminating PM prematurely before evidence supports the change.

Expert Insights to Add Before Publication

• Reliability engineer review of failure-mode selection and P-F interval.
• Approved Titan workflow showing how an alert becomes planned work.
• A de-identified pilot example with baseline, assumptions and limitations—only if measured.
• IT/OT security review of integration guidance.

Frequently Asked Questions

Is predictive maintenance better than preventive maintenance?

It is better only for failure modes where degradation can be detected with useful warning and the economics justify monitoring. Preventive, condition-based, predictive, run-to-failure and redesign can all be correct.

Does predictive maintenance require AI?

No. Inspections, thresholds, trending and statistical techniques can be predictive or condition-based. Use the simplest method that produces a reliable decision.

What is the first step?

Identify critical assets and failure modes, assess current data and maintenance discipline, and choose a bounded use case with a measurable baseline and clear response.

Can a CMMS perform predictive maintenance?

A CMMS can hold asset/work history, schedule inspections and turn decisions into work. Some products may include condition or analytics features. Verify the actual product, data and integration capabilities rather than relying on the category label.

How much data is needed?

It depends on the method, variability and failure rarity. Engineering rules may require less history than a supervised model. The data must represent operating conditions and include reliable outcomes.

What metrics should a pilot use?

Warning time, confirmed-alert rate, false-positive burden, missed failures, action rate, avoided/reduced consequence, program cost and user trust. Model accuracy alone is insufficient.

Titan MMS, CMMS Implementation, Reduce Downtime, Maintenance KPIs, Manufacturing Software, IT/OT Integration and AI Readiness.

Use primary reliability/asset-management standards and public technical resources from recognized condition-monitoring institutions. Cite current NIST guidance for industrial security when discussing connectivity. Avoid vendor ROI claims as universal benchmarks.

Conclusion and CTA

The choice is not preventive or predictive for the whole plant. It is the right policy for each failure mode. Begin with function, consequence, detectability, warning time, action capacity and economics. Build reliable maintenance data and work discipline, then add condition and predictive techniques where they improve a real decision.

The assessment should end with a failure-mode portfolio: which risks stay on preventive schedules, which move to condition routes, which justify a predictive pilot, which run to failure, and which require redesign. That portfolio is more useful than a site-wide slogan about becoming “predictive.”

• Images: technician condition inspection; approved monitoring/work-order screen.
• Diagrams: P-F curve; alert-to-work feedback loop; hybrid strategy map.
• Infographic: seven questions for selecting maintenance policy.
• Tables: failure-mode decision worksheet; pilot evaluation scorecard.
• Comparison chart: preventive, condition-based, predictive, run-to-failure and redesign.
• Video: reliability expert walks through a pump failure-mode example.
• Downloadable lead magnet: maintenance strategy selection workbook.
• Suggested case study link: approved manufacturing/Titan example.
• Suggested product link: Titan MMS.
• Suggested related articles: CMMS Implementation; Downtime Reduction; Maintenance KPIs; Manufacturing AI Readiness; CMMS vs EAM.