Shock Detection

You may want to mount the accelerometer on rubber feet or use some other mechanical arrangement to remove the high-frequency components from the shock. This would help to spread the shock out over a longer time for more reliable detection of a sharp (short duration shock).

40 Hz seems reasonable, but I'm not a mechanical engineer. It depends a lot on the total environment the accelerometer is in. For example, if the accelerometer is already on a PCB that might flex ...

The only observation I'd make about your algorithm is that you might want to go to a rolling window rather than one you reset. You don't need to sum the

60 samples every time--just add in the new sample and subtract out the oldest one, i.e.

int queue[60]; int index; int window_sum;

...

void stuff_new_accel(int new_accel) { int old_index = index;

index++; if (index >= 60) index = 0; queue[index] = new_accel;

sum += new_accel; sum -= (queue[old_index]); }

The Lizard

Define "shock"? What kind of shocks are you looking to detect? Does your algorithm need to reject vibrations?

I suggest you start by capturing a large number of events - real "shock" events and other random events that you think your device will see in real life - and looking at the waveforms. Run some test algorithms over these captured data sets and tweak your catch/reject rates. Then move over to real hardware.

Thank you Jujitsu and Larwe, very helpful hint. To determine what a shock is, I'll do a series of test and recored energy level and fine tunning the alarm threshold. May be I should make the size of the sample buffer a variable to adjust shock period.

e

I had a similar sort of application - which I will not discuss in detail, sorry - and what I did was to make a logger pod out of a Logomatic datalogger and 512MB SD card. I am using an analog accelerometer. I put the "logger box" into a jiffy box and installed it in various representative situations to gather background noise and decide on my algorithm and sense criteria. I sampled at

500Hz on all three axes (note that I had an RC filter with a 303Hz

-3dB point).

shock

the

Furthermore, to work out the shock, I need to work out the vector of energy? sqr(x^2+y^2+z^2) ?

That simply gives you the magnitude of the force vector. I would not suggest reducing it to this scalar immediately because you're hiding important data. To find a shock, one approach is to look at the first derivative of the input signal. I'd suggest that you try sampling into a circular buffer, take the first derivative of the x,y and z curves, combine them to get a magnitude using the Pythagorean formula you quote above, then look for local maxima.

That's not presented to you as a solution, merely as something you can try to help visualize the problem. I don't remember exactly how to use the ST accelerometers in digital mode - I use AD and Kionix parts mainly - but you might need to do some digital filtering to get a good result. Experiment, experiment.