Several solutions to this...
For mission-critical, normal solution is Design Redundancy:
o Serial Inline Fans -- 1 fails, other keeps running
---- if you put one as intake, it will experience cooler air, so last longer
---- Delta have static post-blade-foil airflow straighteners to reduce noise
o Serial Inline Temp Controlled Fans -- 1 fails, other speeds up re temp
---- if both working, they run at a lower rpm, so last longer
---- ex - EBM-Papst 8412MV offer 3rd wire remote thermistor control
--------- remote sensors allow themselves to be sited in exhaust (hot)
--------- whilst the fans can be sited in serial intake for longer life (cooler)
Whether you use a single/parallel or serial fan solution, you can...
o Run fans at a lower rpm than spec - they will last longer
---- NMB have curve data for the effect of rpm as well as temp on L10 data
o Ideal is to run fans at 70-80% of maximum - they will last longer
---- whilst maintaining the chosen fan under such in the sweet-spot of the P-Q curve
o Run fans in air intake mode - they will last longer
---- ambient v exhaust temp delta is quite considerable in some applications
---- particularly where mission-critical involves a HVAC/plenum installation
For mission-critical data systems I strongly prefer redundancy for 2 reasons:
o Redundancy allows you to choose the downtime - not the machine
---- eg, scheduled to off-peak hours based on baseline user load
o Servicing machines to replace fans involves humans & so human error
---- a notable %age of coloco downtime is caused by human error
---- either in servicing the affected machine, or in doing so off-lining others
So redundancy like say RAID-1 is about availability:
o L10 is just an expected figure based on an op-voltage/rpm & air temp
---- typically when 50% of the oil (by mass) in the bearing has oxidised
o In a data Coloco HVAC-maintained environment, the data is useful
---- altho height of air intake from the floor affects intake-temp
---- along with HVAC plenum design/velocity & rack positioning
o Deviate into non-HVAC environments, and mission-critical gets interesting
---- you get wider temperature swings & variability in user implementation
---- ignoring Management sticking the latest ABC report in the fluorinert tank
Some of the high-availability Backbone routers use fans on both intake & exhaust,
and will function with one operational - usually the exhaust fails before the intake.
So if there is space, and cost isn't prohibitive, serial redundant fans works well.
There are some 200k-hr fans out there, I think Sanyo has one and NMB.
They may, however, not be commonly available - ie, OEM 50k-min order.
An increasingly typical solution, is to outsource the problem to the PSU:
o PSUs tend to have finite lives, relatively low MTBFs
---- depends on the design, cooling, and electrolytic capacitor ESR etc
o Typical PSUs can have an MTBF similar to typical fans
---- and so making it a field-replaceable unit is common
o Very high MTBF PSUs often uses redundancy to achieve the figures
---- and additionally to allow dual-power-feed for that redundancy
However, if you are using a higher-grade industrial PSU (eg, 1M-hr MTBF)
then the fans life becomes more important in relation to that unless PSU derated.
For tacho signal outputs:
o DC fans produce a switching noise which can be used to work out tacho
---- that is for fans who directly offer no tacho output
o EBM-Papst fans with tacho have the code extension /2 on the end
---- eg, an 8412 NGML with tacho becomes 8412 NGML/2
o NMB fans with tacho are generally to order only
---- they require the B noise level code to end in a 9, eg, B59
---- AND the A option code must also be 50-100, eg, B59-A100
---- so a B59-A10 despite the 9 would be a fan without tacho signal
An alternative signal output is locked rotor:
o This is a signal which goes high (typically) when the rotor stops (locked)
o Such a condition would be when the fan is (well
beyond service life
o Typically Panaflo 1A use this method by default, tacho is by code 1BX
---- DEC Alpha, SUN & SGI use the 1A method
---- General PCs use the 1BX method - although there is some overlap
Locked-Rotor can be cheaper to support re logic, although lesser if the
mission-critical solution needs more involved reporting & notification as
there are specific chip options to implement tacho fan monitoring/alarm.
Contacting the major fan manufacturers with your requirements may get
them to do the legwork for you in finding a solution to meet your needs.
This is particularly true of NMB & EBM-Papst engineers.
If the design of the fan cooling system isn't closed (and it should be done in
the earliest stages re cost) then I still favour redundancy for mission-critical.
If only, physical redundancy allows you to choose the downtime. You can
pick a super-high 1.2M-hr MTBF Seagate drive, some still fail far earlier
and it's that servicing downtime and inability to choose it that annoys/costs.
Talking of which, I'm off to chase a low MTBF failure on a Maxtor
Hope that helps, group helped me before so reciprocating.
PS. Noisier/higher-flow fans are in production with tacho, hunt around.
It's the very low noise ones which aren't commonly in production, usual
reason since it costs an integrator 0p to make 1 fan go faster, but costs
them quite a lot to use 2 fans re lost profit & perhaps low mkting benefit.