RECAP | ROUNDTABLE | Effective Predictive Maintenance Program

Introduction:
Predictive maintenance has been a routine data center practice for many years. However, there seems to be confusion and deficiencies with many data center operators as to what constitutes an effective predictive maintenance program. Juan Orozco, Uptime Institute consultant, joins a roundtable discussion that will allow attendees to engage and interact on basic predictive maintenance best practices. What are the basic pieces of equipment that are utilizing predictive maintenance? What takes predictive maintenance from a serious of scheduled tasks to a fully effective sustainable program?

The group started the discussion by indicating what they perform in the way of predictive maintenance activities.
• Infrared (IR) scanning, mostly perform annually
• Battery monitoring systems used, supplemented by manual maintenance performed quarterly by most
• Engine oil analysis, some sample quarterly, most semi-annually or annually depending on how frequent the engines operated
• Refrigerant sampling where applicable
• Vibration testing on rotating equipment (fans, motors, cooling tower fans, engines, air handlers), performed annually, some semi-annually - It was indicated this type of predictive maintenance is more high-end and not usually performed at most data centers
• Fuel testing, some perform quarterly, most perform annually - Some have fuel polishing systems as well
• Eddy current testing, 2 years for condenser, 4 years for chiller
• Ultrasonic testing used for rotating equipment with fluid moving through it - This testing is not all that common in data centers and also consider more high-end

It was mentioned how monitoring power quality is important. This is typically measured already by switchgear, circuit breakers, UPS, etc., and provides harmonics and power factor readings. Monitoring ground resistance can be important as well. The comment was raised to consider the value versus the cost of carrying out this type of maintenance. Once the building is commissioned, having ground resistance retested on a frequency is not something normally done.

An attendee stated air quality sampling is typically performed in data halls located in the UK. The testing looks for small particulates that could impact the IT hardware, using a strip of copper and silver to catch particulates that could lead to potential degradation. The folks in North America commented how this type of testing is not typically done in North America.

Now that you have all this predictive maintenance data, what do you do with it?
• Performance trending
• Analyze for potential premature failures mostly
• Make fact-base decisions in regard to maintenance on this equipment

It was mentioned how equipment vendors are mostly utilized to perform this data collection and trending analysis, which is a potential issue because this means there is a reliance on these vendors. It would be preferable if the equipment vendors provided this data to onsite staff and they were knowledgeable and trained to do the data review and analysis.

An attendee asked about how prevalent vibration analysis is in data centers. The response was that vibration analysis is not that prevalent for the most part, primarily it seems because of cost versus benefit. However, as was brought up, if it can avoid or find a potential engine failure, wouldn’t that justify the expense?

To summarize, turn predictive maintenance into a program by documenting the process, what’s being done on what equipment, and what you then do with the data from a trending and analysis perspective. The basic predictive maintenance tasks all should be doing, if applicable to your data center:
• IR scanning
• Oil sampling and testing
• Fuel sampling and testing
• Refrigerant sampling and testing
• For batteries, utilize battery monitoring systems coupled with regular onsite maintenance
• Eddy current testing for chillers
More advanced predictive maintenance items would be vibration and ultrasonic testing. Consider cost versus benefit when making decisions on what predictive maintenance to perform. Lastly, predictive maintenance must be part of your continuous training program in order to be effectively implemented and managed.