Dynamic Maintenance Cost Optimization in Data Centers: An Availability-Based Approach for K-out-of-N Systems

Overview

The study asks how to link UPS asset condition to reliability and meet a target availability at the lowest maintenance cost. The method is dynamic programming: a ten-condition-state table maps each UPS's condition to MTBF, MTTR and availability, and the model chooses which units to maintain each month. The case study uses ten 500 kW modular UPS units, with maintenance service options and prices based on the APC Symmetra PX 500 kW UPS.

The condition-state table runs from brand new (MTBF 933,708 h, MTTR 2 h, availability 99.999%) and fully operational (850,000 h, 8 h) through fair (100,000 h, 18 h, 99.982%) and poor (50,000 h, 25 h, 99.950%) to critical near end of life (10,000 h, 30 h, 99.701%). In the first-month case, five of the ten UPS units were selected for maintenance, at a monthly budget of USD 9,973 and a combined system availability of 99.991%, above the Tier I requirement of 99.671%.

Optimal UPS count and minimum monthly maintenance cost by availability target: 99.671% (Tier I) 5 units, USD 9,973; 99.741% (Tier II) 6 units, USD 11,417; 99.982% (Tier III) 8 units, USD 14,856; 99.995% (Tier IV) 10 units, USD 19,946. A sensitivity study varying MTBF, MTTR and cost by ±10% shows that raising the maintenance budget does not always buy proportional availability.

Limits: the model is tied to one DC configuration (k-out-of-n UPS) and cannot be applied directly to other subsystems such as HVAC or network; failure rates and maintenance costs are assumed because real operating data is variable and scarce. The authors suggest a multi-component extension covering items such as cooling, and a modular version with subsystem-specific parameters.

Key points

1. Condition-state table: brand new MTBF 933,708 h, MTTR 2 h (99.999%); fully operational 850,000 h, 8 h; fair 100,000 h, 18 h (99.982%); poor 50,000 h, 25 h (99.950%); critical 10,000 h, 30 h (99.701%).
2. In the first-month case, five of ten UPS units were selected for maintenance at a monthly budget of USD 9,973 and a combined system availability of 99.991%.
3. Optimal UPS count and minimum monthly cost by target: 99.671% (Tier I) 5 units, USD 9,973; 99.741% (Tier II) 6 units, USD 11,417; 99.982% (Tier III) 8 units, USD 14,856; 99.995% (Tier IV) 10 units, USD 19,946.
4. Maintenance service options and costs are based on the APC Symmetra PX 500 kW UPS; the case has ten 500 kW modular UPS units.
5. The ±10% sensitivity study on MTBF, MTTR and cost showed that a bigger maintenance budget does not always give proportional availability gains.
6. The model uses assumed failure rates and maintenance costs because real operating data is limited.

Using it in design

• Plan UPS maintenance from a condition table that maps each unit's state to MTBF and MTTR, then pick the units to maintain that meet target availability at the lowest cost.
• Report the minimum number of active units and the maintenance cost together for each Tier target (for example 5 units for Tier I, 10 units for Tier IV).
• Show the effect of maintenance budget on availability through sensitivity analysis and identify the point of diminishing returns.
• Replace the assumed failure-rate and cost inputs with operating records as they become available.

Who should read it

Facility managers and reliability engineers planning UPS maintenance and balancing redundancy against cost.

Document contents

1. 1. Introduction (1.1 Literature Review, 1.2 Research Gap Analysis, 1.3 Research Significance)
2. 2. Materials and Methods (2.1 Problem Formulation and Algorithm, 2.1.1 Model's Parameters, 2.1.2 Dynamic Programming Approach, 2.2 Case Study Implementation, 2.2.1 System and Component Description, 2.2.2 Failure Modes and Condition States, 2.2.3 Reliability and Availability Information, 2.2.4 Dynamic Failure Rates and Maintenance Costs)
3. 3. Results
4. 4. Discussions (4.2 Key Limitations and Constraints)
5. 5. Conclusions

Related standards

• Uptime Institute Tier Standard: Topology

Read the document

Publication details

Authors

Mostafa Fadaeefath Abadi; Mohammad Javad Bordbari; Fariborz Haghighat; Fuzhan Nasiri

Publisher

Buildings (MDPI AG)

Year

2025

Volume

15

Issue

7

Pages

1057

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.3390/buildings15071057

Licence

CC BY 4.0 — share with attribution

Licence evidence

www.mdpi.com

Citation