tcmtech and Sceadwian - your right the whole topic of condition monitoring is fascinating and for many years our company fitted monitoring systems to jet engines, power stations, petrochem plants etc etc so this side of the equation is my background. The electronics side is a hobby and used as a way to solve a need.
The vibration you both mention is correct and mainly used in larger pump installations but for my application it cannot be used. The pumps are small and mounted on large noisy machinery that would mask any signal - certainly the small changes that need to be detected to signal any impending failure.
Some of the pumps under study are magnetic drives, that is they have no bearings at all and float inside a circular rotating magnet, this in turn precludes many types of sensors due to the rotating field.
Other pumps are ceramic or ceramic tipped and a chipped ceramic tip will lose performance but the imbalance could be small as the tip weight is minimal compared to the whole disc assembly.
The type of sensors that can be used need to be 'good value' so accelerometers suitable for this are too expensive.
Piezo discs, strain gauge bridges can be bonded to the casing to get a signal but the processing is labourious and difficult because of the background noise.
The best way to do this is really to collect a 256 or 512 block of data from an ADC subsystem (with AA filters etc) and perform an FFT on it, the peak of the main frequency will change height and eventually will change frequency to 5/6th of the original value. Tracking the frequency with an adaptive filter would be easy and reliable in its detection rate. There are many microchips that do this ie TI C20 or C40 ....But its a lot of kit, expensive, complicated and I would not able to stop myself from including all sorts of extra stuff to detect other failures, extra sensors, multiplexed sensors etc it would end up being a huge project.
Also I mentioned earlier that differant pumps run at differant speeds and also have up to 20 blades so the 1 in 6 being discussed is just for one pump.
At each location there could be anywhere from 1 to say 7 or 8 pumps that each need an independant detector so the only way I can think of a cheap, simple and small detecter that works in all scenarios is a low peak detector.
Any other ideas seriously considered. (Colin has already got me doing a course on PIC's!!!!)
Quantised