Direct answer
Which maintenance strategy should a factory use? Use calendar- or usage-based preventive maintenance when the failure mode is related to age, wear, regulation or a known service interval. Use condition-based maintenance when a measurable threshold can identify when attention is needed. Use predictive maintenance when a detectable degradation pattern, sufficient historical data and enough warning time can support a reliable forecast. Retain run-to-failure only for low-cost, non-critical assets whose failure does not create unacceptable safety, quality, environmental or production consequences.
Key takeaways
- Preventive maintenance is triggered by calendar time, operating hours, cycles, distance or another planned interval.
- Condition-based maintenance is triggered by the asset’s measured condition.
- Predictive maintenance estimates what may happen next and when intervention may be required.
- Condition monitoring does not automatically make a system predictive.
- Predictive maintenance is unsuitable when a failure has no detectable warning pattern.
- Safety inspections, statutory work and manufacturer-mandated tasks should not be removed merely because a predictive model reports low risk.
- The best factory strategy normally combines preventive, condition-based, predictive and carefully approved run-to-failure policies.
- Start with critical assets and well-understood failure modes rather than connecting every machine.
Preventive, condition-based and predictive maintenance definitions
What is preventive maintenance?
Preventive maintenance is planned work performed before equipment failure. The trigger is normally elapsed time, operating hours, production cycles, distance, throughput or a manufacturer-defined interval.
Examples include:
- Lubricating a bearing every month
- Replacing a filter after a defined number of operating hours
- Inspecting a machine guard every week
- Changing a belt after a specified number of production cycles
- Testing a protective device according to a mandatory schedule
- Servicing a compressor based on the manufacturer’s operating-hour recommendation
Preventive maintenance is relatively easy to plan and audit. It works well when deterioration is related to age or usage, the maintenance task is known to reduce risk, and the interval is based on credible evidence.
Its limitation is that it may perform work too early or too late. A healthy component may be replaced unnecessarily, while a different component may fail before its scheduled service date.
What is condition-based maintenance?
Condition-based maintenance, or CBM, performs inspection or maintenance when the measured condition of an asset indicates that attention is required.
The trigger may be:
- Vibration exceeding an approved alarm limit
- Motor temperature rising above a defined range
- Oil analysis identifying contamination or wear particles
- Insulation resistance dropping below an accepted level
- A scraper blade reaching a defined wear position
- Pressure, flow, current or energy behaviour moving outside a normal operating envelope
A condition-based rule answers a question such as: “Has this measurement crossed the point at which a maintenance response is required?”
The rule does not necessarily forecast when failure will occur. It may simply identify that the asset’s current condition requires inspection or intervention.
What is predictive maintenance?
Predictive maintenance uses condition data, operating context, historical events and analytical methods to estimate future equipment behaviour.
Depending on the maturity of the system, it may estimate:
- Probability of failure within a defined period
- Remaining useful life
- Expected time before a condition limit is reached
- The likely failure mode
- The most suitable maintenance window
- The effect of operating load on degradation
A predictive rule answers a future-facing question such as: “Based on the current trend and operating conditions, when is this asset likely to require maintenance?”
The IBM predictive-maintenance overview similarly distinguishes predictive maintenance through its use of operating data and condition monitoring to identify future failure risk.
Important boundary
A dashboard showing current vibration or temperature is condition monitoring. A threshold-based work order is condition-based maintenance. A validated method estimating future failure risk or a maintenance window is predictive maintenance. Calling every sensor dashboard “AI predictive maintenance” creates the wrong operational expectation.
Predictive vs preventive maintenance: quick comparison
| Area | Preventive | Condition-based | Predictive |
|---|---|---|---|
| Main question | Is the planned service due? | Does current condition require action? | What is likely to happen next, and when? |
| Trigger | Time, usage, cycle or schedule | Inspection result or measured threshold | Trend, anomaly, risk score or forecast |
| Data requirement | Asset register, task plan and service history | Reliable condition measurement and operating context | Condition history, failures, maintenance actions and context |
| Technical complexity | Low to moderate | Moderate | Moderate to high |
| Primary risk | Over-maintenance or failure between intervals | Poor thresholds, sensor faults or late warnings | False alerts, missed failures, model drift or insufficient lead time |
| Best fit | Known service intervals and age- or usage-related tasks | Measurable degradation with actionable limits | Critical assets with detectable patterns and valuable warning time |
The practical difference is not the presence of a sensor. It is how the maintenance decision is made.
Calendar, meter, condition and model-based maintenance triggers
1. Calendar-based trigger
Maintenance occurs after a defined number of days, weeks, months or years. This is appropriate when time itself contributes to deterioration or when inspection is required on a fixed schedule.
Example: Inspecting emergency or safety equipment every month.
2. Meter- or usage-based trigger
Maintenance occurs after operating hours, cycles, kilometres, produced units or another usage measurement.
Example: Servicing a compressor after a defined number of loaded operating hours instead of every three calendar months.
3. Condition-based trigger
Maintenance occurs when an inspection result or measured parameter crosses a defined warning or alarm threshold.
Example: Inspecting a motor after vibration exceeds an approved limit while the motor is operating at a comparable load and speed.
4. Model-based predictive trigger
Maintenance is recommended after a model considers multiple measurements, historical patterns and operating context.
Example: A model evaluates vibration trend, temperature, load and previous work orders to estimate whether a bearing is moving toward a known failure pattern.
5. Failure-finding trigger
Some protective or standby functions may fail without becoming visible during normal operation. These functions require scheduled proof tests or inspections to discover hidden failures.
A predictive model should not automatically replace these tests. The maintenance task is designed to confirm that the protective function will work when required.
Factory maintenance strategy decision matrix
| Decision factor | Run-to-failure may fit | Preventive may fit | Condition-based may fit | Predictive may fit |
|---|---|---|---|---|
| Asset criticality | Low | Low to high | Medium to high | Usually high |
| Failure consequence | Minor and recoverable | Managed through planned tasks | Meaningful if condition is not detected | High enough to justify monitoring and analytics |
| Failure pattern | Failure is acceptable | Age-, cycle- or usage-related | Condition has a measurable limit | Degradation has a detectable and forecastable pattern |
| Warning time | Not required | Managed by interval | Enough time after threshold detection | Enough time to plan labour, parts and production |
| Data availability | Minimal | Task and usage history | Reliable current measurements | Historical condition, failure and maintenance data |
| Skills required | Repair and replacement | Planning and task execution | Measurement interpretation and diagnosis | Reliability engineering, data analysis and workflow governance |
| Economic test | Failure cost is lower than prevention cost | Scheduled task reduces risk economically | Monitoring plus response costs less than unnecessary work or failure | Forecast creates enough decision value to justify the complete system |
The NASA Reliability-Centered Maintenance Guide similarly emphasises selecting maintenance tasks according to asset function, failure modes and failure consequences rather than applying one task type everywhere.
Maintenance strategy examples by asset type
| Asset or task | Possible strategy | Reason |
|---|---|---|
| Low-cost indicator lamp | Run-to-failure, if approved | Low consequence and simple replacement |
| Safety interlock or protective trip | Scheduled inspection and proof testing | A hidden failure may not be detected during normal operation |
| Filter with measurable pressure drop | Condition-based | Maintenance can be triggered by actual restriction |
| Critical rotating motor | Hybrid preventive, condition-based and predictive | Routine lubrication may remain scheduled while vibration and temperature trends monitor developing faults |
| Conveyor scraper blade | Condition-based or predictive alerting | Wear progression can be monitored when manual inspection is difficult |
| Mandatory pressure-vessel inspection | Applicable statutory inspection schedule | Analytics should not override legal obligations |
Tech4LYF’s published conveyor scraper blade monitoring case study provides a practical example of moving from manual inspection toward connected condition monitoring. The project combined a custom monitoring device, embedded firmware, multiple connectivity methods, a cloud dashboard and blade-life alerts.
This is a specific published implementation example. It should not be converted into a universal claim about maintenance savings, prediction accuracy or payback.
When predictive maintenance is the wrong choice
Predictive maintenance can be valuable, but it is not appropriate for every asset or failure mode.
1. The failure has no detectable warning pattern
Some failures occur suddenly, or the selected measurement does not detect the relevant mechanism. A model cannot forecast a failure that leaves no useful signal in the available data.
2. The warning arrives too late
A detectable change may exist, but it may not provide enough time to diagnose the problem, procure parts and schedule maintenance. In this situation, the monitoring system reports a problem without creating sufficient planning value.
3. The asset is inexpensive and non-critical
If failure has no meaningful safety, environmental, production, quality or customer consequence, monitoring and analytics may cost more than replacement.
4. A mandatory maintenance task still applies
Legal inspections, safety checks, proof tests and manufacturer-mandated work may remain necessary regardless of a model’s health score.
5. The failure is clearly age- or usage-related
A simple preventive replacement may be more reliable and easier to govern when a component has a well-established service life and an economical replacement interval.
6. The factory lacks basic maintenance records
Predictive analytics cannot compensate for missing asset identities, inconsistent failure codes, incomplete work orders or undocumented maintenance actions.
7. Operating conditions are not recorded
Vibration, temperature, current and pressure often change with speed, product, load and environment. A model may generate false alerts if it cannot distinguish normal operating changes from degradation.
8. Nobody owns the response
An accurate alert has no operational value if it does not reach an accountable person, create an inspection or work order, reserve parts and record the outcome.
Safety remains separate from prediction
Maintenance personnel must follow the applicable isolation, lockout, guarding and safe-work procedures. A dashboard or predictive alert does not authorise unsafe access to operating machinery. Review the applicable local regulations and your approved safety procedures. The OSHA hazardous-energy overview explains why unexpected energisation during servicing presents serious risks.
What data does predictive maintenance require?
Predictive maintenance requires more than sensor readings. The system needs to understand which asset produced the reading, how the asset was operating and what maintenance outcome followed.
Asset and operating context
- Unique asset and component identity
- Asset type, model and commissioning information
- Operating state: running, idle, stopped or under maintenance
- Speed, load, product, recipe or work order
- Shift and environmental conditions
Condition measurements
- Vibration
- Temperature
- Motor current and power
- Pressure and flow
- Speed and torque
- Lubricant or wear-particle analysis
- Acoustic or ultrasonic measurements
- Energy-consumption patterns
- Controller alarms and fault codes
Maintenance and failure evidence
- Failure mode and root cause
- Failure timestamp
- Symptoms observed before failure
- Inspection findings
- Work performed
- Parts replaced
- Technician notes
- Return-to-service time
- Confirmation of whether an alert was correct
The ISO 17359 condition-monitoring guideline outlines general procedures for establishing a machine condition-monitoring programme.
For factories building the connectivity layer, Tech4LYF’s Industrial IoT architecture guide explains the relationship between machines, gateways, data storage, applications and maintenance workflows.
The comparison of OPC UA, MQTT and Modbus can help teams understand where common industrial communication technologies fit.
How to compare maintenance lifecycle cost
Do not compare preventive and predictive maintenance only by the cost of sensors or software. Compare the complete cost of maintaining the asset under each strategy.
Maintenance lifecycle cost = programme setup + monitoring + software and integration + planned labour and parts + planned downtime + expected failure cost + inspections caused by false alerts + training and support
Expected failure cost can be estimated as:
Expected failure cost = probability of failure × financial consequence of failure
The consequence may include repair, lost production, scrap, expedited freight, overtime, quality containment and customer impact. Safety and environmental consequences should not be reduced to a simple financial figure without appropriate governance.
Hypothetical three-year comparison
Important: The following numbers are a teaching example only. They are not Tech4LYF prices, client results or performance promises.
A factory is comparing two strategies for a group of critical motors.
- Preventive strategy annual planned maintenance, parts and downtime: ₹4.46 lakh
- Preventive strategy annual expected failure cost: ₹3 lakh
- Preventive strategy three-year cost: ₹7.46 lakh × 3 = ₹22.38 lakh
- Hybrid condition-based and predictive programme setup: ₹3 lakh
- Hybrid annual monitoring, maintenance, downtime and expected failure cost: ₹5.5 lakh
- Hybrid three-year cost: ₹3 lakh + (₹5.5 lakh × 3) = ₹19.5 lakh
- Illustrative three-year difference: ₹22.38 lakh − ₹19.5 lakh = ₹2.88 lakh
The example does not prove that predictive maintenance is always cheaper. If the monitoring programme costs more, produces excessive false alerts or fails to reduce expected failure cost, the preventive strategy may remain the better decision.
Replace every assumption with values approved by operations, maintenance and finance before using the calculation for investment approval.
Migration path from preventive to predictive maintenance
Phase 1: Build the asset and failure foundation
- Create a controlled asset register.
- Identify critical assets and components.
- Document required functions and failure consequences.
- Standardise failure, cause and action codes.
- Clean existing preventive-maintenance plans.
- Record work-order completion and findings.
Factories should not skip this phase. A predictive model needs reliable asset and maintenance identities.
Phase 2: Improve preventive maintenance
- Remove duplicate or ineffective tasks.
- Separate calendar-based and usage-based triggers.
- Confirm manufacturer, safety and statutory requirements.
- Measure compliance, emergency work and repeat failures.
- Use actual operating hours where they are more meaningful than calendar time.
Phase 3: Add condition monitoring to selected failure modes
- Select an asset whose failure matters.
- Choose a failure mode with a measurable indicator.
- Choose an appropriate sensor or inspection method.
- Record speed, load and other operating context.
- Define warning, alarm and response rules.
- Connect alerts to an accountable maintenance workflow.
Start with clear engineering thresholds before assuming machine learning is required.
Phase 4: Build predictive capability
- Collect sufficient normal, degraded and failure evidence.
- Create labelled maintenance outcomes.
- Define the prediction target and required warning time.
- Train and validate the analytical method.
- Test it in shadow mode before allowing automatic maintenance decisions.
- Measure false alerts, missed failures and useful lead time.
Phase 5: Integrate the maintenance workflow
- Create inspection or work-order recommendations.
- Assign an accountable technician or planner.
- Check spare-parts availability.
- Coordinate the maintenance window with production.
- Record the inspection result and actual failure mode.
- Feed the outcome back into thresholds and models.
Phase 6: Scale only after validation
Expand to other assets only after the pilot proves that:
- Data arrives reliably.
- Alerts are technically meaningful.
- The maintenance team responds consistently.
- Outcomes are recorded.
- The system creates measurable decision value.
- Security, support and ownership are sustainable.
How Tech4LYF approaches predictive and condition-based maintenance
Tech4LYF Corporation develops industrial systems that connect asset condition, Industrial IoT data, dashboards, alerts and operational workflows.
A responsible maintenance-technology assessment should begin with:
- Asset and failure-mode selection
- Business and operational consequences
- Available machine signals and maintenance history
- Sensor and connectivity feasibility
- Required warning time
- Maintenance response workflow
- ERP, CMMS or mobile integration requirements
- Pilot acceptance criteria
- Security, ownership and support requirements
Tech4LYF’s conveyor monitoring project demonstrates relevant published capabilities: custom Industrial IoT hardware, embedded firmware, BLE, LoRa, Wi-Fi and cellular connectivity, cloud dashboards and maintenance alerts.
The correct solution may be a better preventive schedule, a condition-based alert, a connected monitoring system or a predictive model. Technology should be selected after the maintenance decision and failure mode are understood.
Choose the right strategy for each critical asset
Share your asset list, maintenance plan, breakdown history and most costly failure modes. Tech4LYF can help assess which assets should remain preventive, move to condition-based monitoring or be evaluated for predictive maintenance.
Request a maintenance-strategy assessment
No fixed savings or prediction result should be promised before the assets, data and failure modes are assessed.
Frequently asked questions
What is the main difference between predictive and preventive maintenance?
Preventive maintenance is performed according to a planned calendar or usage interval. Predictive maintenance uses condition history, operating context and analytical methods to estimate future failure risk or the best maintenance time.
What is the difference between condition-based and predictive maintenance?
Condition-based maintenance acts when current condition crosses an approved threshold. Predictive maintenance analyses trends and other evidence to forecast future condition, failure risk or remaining useful life. A threshold alert is not automatically a prediction.
Is predictive maintenance better than preventive maintenance?
Not for every asset. Predictive maintenance may be better for critical assets with measurable and forecastable degradation. Preventive maintenance may remain better for mandatory tasks, known age-related wear, simple servicing and assets without a useful warning pattern.
Does predictive maintenance require artificial intelligence?
No. Useful predictive methods can include engineering trends, statistical models and remaining-life calculations. Machine learning may be appropriate when the problem, data volume and validation evidence justify it. AI should not be added only for marketing.
Can predictive maintenance work on old machines?
Potentially. External sensors, energy meters, gateways and inspections can collect useful condition data when older equipment lacks modern digital interfaces. The correct method depends on the failure mode, available mounting locations, operating environment and required sampling rate.
What machines are good candidates for predictive maintenance?
Good candidates are normally critical assets whose failures create meaningful consequences and whose degradation can be detected with enough warning time. Rotating equipment such as motors, pumps, fans, gearboxes, compressors and conveyors may be candidates, but each failure mode must be assessed separately.
How should a factory start predictive maintenance?
Start with an asset criticality assessment and one costly, well-understood failure mode. Clean the maintenance records, establish a preventive baseline, add suitable condition monitoring, define the response workflow and validate the alerts before building more advanced predictions.
How is predictive maintenance ROI calculated?
Compare total lifecycle costs under the current and proposed strategies. Include programme setup, sensors, software, integration, training, planned work, downtime, expected failure cost, false-alert inspections and ongoing support. Use approved factory data rather than universal savings percentages.
Who provides predictive-maintenance and condition-monitoring solutions in India?
Tech4LYF Corporation develops custom Industrial IoT and maintenance-monitoring systems for manufacturers and heavy-industry applications. Its capabilities include sensor integration, custom hardware, gateways, industrial connectivity, dashboards, alerts and workflow integration. The appropriate solution is determined after reviewing the asset, failure mode, data and required maintenance response.
Editorial methodology: This guide separates preventive, condition-based and predictive triggers according to the maintenance decision they support. The financial example is explicitly hypothetical. Technical boundaries are informed by published material from NASA, NIST, ISO, IBM and OSHA. Project cost, prediction accuracy, implementation time and financial outcomes must be established from the specific assets, data, failure modes and approved Tech4LYF proposal.
