Maintenance in a data center should preserve reliability without creating unnecessary operational risk. Performing too little maintenance allows deterioration to accumulate, while performing intrusive work too frequently can itself increase the probability of human error or equipment failure. The right strategy is risk-based and evidence-driven.
Classify assets by criticality
Not every asset deserves the same maintenance strategy. A main UPS, generator, chilled-water pump or critical switchboard has a different operational consequence from noncritical lighting equipment. Asset criticality should consider failure impact, redundancy, replacement time, detectability and safety risk.
Preventive maintenance
Preventive maintenance is performed at planned intervals based on time, operating hours or manufacturer recommendations. It can include inspection, cleaning, lubrication, torque verification, filter replacement, functional testing and component replacement.
Manufacturer instructions should remain a primary technical basis for equipment-specific maintenance.
Predictive and condition-based maintenance
Predictive maintenance uses measured condition to identify deterioration before failure. Examples include thermal imaging, vibration analysis, battery impedance trends, oil analysis, transformer temperature trends, breaker operation counts and electrical power-quality data.
Condition-based maintenance can reduce unnecessary intervention when reliable condition indicators are available.
Build the maintenance plan from the failure mode
A useful maintenance task should detect, prevent or mitigate a credible failure mode. If a task cannot explain which risk it controls, its value should be questioned.
For example, battery inspection addresses connection, temperature and physical-condition risks, while load testing provides evidence of functional capacity.
Maintenance windows
Before starting work, determine the resulting system state. Will N+1 become N? Will one complete A or B path be unavailable? What happens if another component fails during the work?
Maintenance windows should therefore consider load level, weather, business events, available redundancy, vendor support and rollback time.
Planned maintenance requires controlled procedures
High-risk maintenance should be performed under an approved MOP with prerequisites, step sequence, hold points, rollback instructions, communication plan and sign-off. Where appropriate, independent verification should be used for critical switching steps.
Use maintenance evidence
A maintenance report should record measured values, findings, defects, parts replaced, alarms tested and corrective actions. “PM completed” alone provides little engineering value.
Historical results should be trended to identify deterioration across repeated maintenance cycles.
Lifecycle replacement planning
Some components have age-sensitive reliability characteristics, including batteries, capacitors, fans, filters, seals and certain control components. Replacement should be planned before risk becomes unacceptable.
ISO 55001:2024 provides requirements for an asset-management system and supports a structured approach to lifecycle decision-making.
Maintenance program review
- Review overdue preventive maintenance.
- Identify repeat failures.
- Trend condition indicators.
- Check whether PM tasks detect useful defects.
- Update intervals using evidence and manufacturer guidance.
- Plan age-sensitive replacement in advance.
- Track defects to closure.
Key takeaway
The best maintenance program is not the one with the most tasks. It is the one that controls the most important failure risks with the least unnecessary intervention. Asset criticality, manufacturer guidance, condition data, disciplined procedures and lifecycle planning should work together to maintain resilience.
References and Further Reading
- ISO 55001:2024, Asset management — Asset management system — Requirements.
- ISO/IEC TS 22237-7:2018, Data centre management and operational information.
- Applicable equipment manufacturer operation and maintenance manuals.
