PIC and fast integration

[shantanu@india]

Hi,
In the design stages of pump vibration monitoring using a 3-axis ADXL345 accelerometer, I am facing the fundamental question of where and how to do the acceleration to velocity conversions? The fundamental mechanical vibration frequency of centrifugal pumps generally do not exceed 60Hz.The output data rate(ODR) of a ADXL345 can be as high as 3200Hz with SPI interface and the raw acceleration values in terms of 'g' has to be integrated to RMS velocity to be of any genuine industrial use.
Now the point really is a PIC 24 / dsPIC the right choice or should I simply transmit the raw values via a PIC to a linux single board computer to do FFT there? I personally would prefer a PIC doing the conversions locally and displaying the real time velocity values on a small module mounted on the pump.
Any ideas?

[top204]

For those times, a standard PIC18F device operating at 64MHz (16MIPS) will do the job of reading and calculating and transmitting or controlling.

Reading the accelerometer takes milliseconds, so multiple readings can be performed for filtering in a fraction of a second (literally). :-)

If 100% accuracy is not required, integer calculations can be performed extremely fast, and even floating point is faster than most people think it is with the Positron8 compiler.

[SCV]

A word of caution, whilst the mems sensors are very small and convenient, the accelerometers detect magnetic flux (gravity) and the gyros use a circle of piezo fingers oscillating in a wave pattern to interact with the coriolis effect.
The magnetometers are of no use near any stray magnetic fields - like one would find on a motor. The piezo fingers can be upset with high frequency vibration giving noisy output. We used them on a helicopter and had to isolate the gyro from the aircraft frame using soft foam pads and by adding inertial mass to the gyro sensor.

You may be better off putting a microphone on your pump and analysing that.
Tim.

[shantanu@india]

Thanks for the tip Tim.
Yes Les , I was also of the opinion that a 8bit can do the job unless some high end analytics are needed.But to convert the acceleration to velocity by numerical integration we need a sample time of at least one second & a sample rate of 10/20 times the fundamental frequency. This for 3-axis floats means a lot of RAM. The device choice becomes an issue.