[familyop]

"I've successfully run 5 - 8 watt PC104 CPU boards on my railcars. It takes very aggressive power management to keep the lead acid storage battery charged from intermittent input from the Timken bearing generator (20 watts max at 15 MPH). I use a PIC microcontroller to manage the charging cycle so the x86 is only alive when necessary."

That's one strange mix of old and new technologies! :-) Are those tiny railcars or heavy ones? ...golf cart type lead acid batteries?

[Myrddin]

The rail cars carry 110,000 lbs of coal. Bearings are class F. One bearing is outfitted with a Timken bearing generator. The batteries are deep cycle marine or gel cell. It is actually a moving magnet, fixed coil alternator (12 poles) that generates up to 260 volts at 80 MPH. The "break even" speed is 10 MPH to generate enough to just power the PC104 and all sensors. The Smart Battery Charger includes a PIC microcontroller to read tachometer pulses off the bearing generator and start the PC104 when it is apparent that the car will be in motion long enough to matter. The PIC reads voltage/current off the generator and battery. That info is fed to the PC104 over a CAN bus. The PC104 sends that data back to the server over a 1xRTT (CDMA) cellular connection to the internet.

There is also a GPS in the box with the PC104 to set time and provide lat/lon information to track the car. I have other PIC controllers measuring bearing temperatures and accelerometers to record longitudinal, lateral and vertical motions. I calculate moving averages for temperature and vibration. That is also sent to the back end server.

The high bandwidth vibration signatures are handled with a Diamond Systems DMM32 to read 100,000 samples per second from the vertical accelerometers on left and right wheels on each axle. About 20 seconds of sample is taken, then subjected to a multi step Fourier analysis. The target frequencies are characteristic of the cone, cup, cage and rollers in the bearing. I pick off the fundamental plus 5 harmonics. That data reduction is sent to the server as well. The end product of the analysis is an early forecast of the kinds of defects occurring in the bearings. The approach can identify 55 specific kinds of defects.

An earlier design used 16 channels of the Diamond DMM32 to measure the bearing temperatures. That is terribly wasteful of the device that retails for $750. I'm accomplishing the same thing with a couple of $10 PIC18F6680 devices and transmitting the results to the PC104 over a CAN bus instead of very expensive low loss analog wiring. The next gen for the high bandwidth analysis will use a stereo audio A to D device driving a dedicated DSP chip to do the bearing analysis. That will replace the $750 card with less than $50 in electronics and do the same job.