Accellerometers and measuring vibration

Greetings from London. I stumbled across your site from Mikeslist.com and was hoping maybe you could help me. First, I must confess that I am not an engineer but rather someone with a few too many goofy ideas. Here’s what I want to do:

I have a property that is near some train tracks and the neighbourhood residents and I are all upset about the one or two freight trains per day that that cause noticeable vibration to the properties. I would love to somehow measure, record, and graph this vibration and publish it on a local web site. That way I can publicise it and put some pressure on the local government.

Even though I am a techie, I know very little about sensors and data acquisition. I have a few older spare PCs to dedicate to the task and can maybe even cobble a few lines of basic C code together if need be.

I need help with:

  • Finding a good, cheap and suitable sensor (I am assuming an accelorometer).

  • Finding the best way to get the data onto a PC and stored in a suitable data format that I can then turn into CVS files or even a graph.

Any suggestions would be most welcome!

Thanks,

Ian.

Hi Ian!

Supposing the vibrations are strong enough to be detected with ±3 milli-G resolution, there is a VERY easy solution to this:

Sparkfun.com has a serial port accelerometer with high precision for 90 USD or lower precision for 50USD. Plug this into the PC serial port, write a simple C program that reads from the serial port and puts the clear text output into one line each in a file. Then import the file to Excel or similar and do what you want with it. This is simplicity itself!

Good luck!

Very cool that you’re using technology and ACTUAL NUMBERS to spur action from the local government. Best of luck and let us know if you need help. It sounds very feasible; if you got crazy, the wi-tilt is just an accelerometer and you could then put them in a few different places and see what effect distance has on the vibration levels… But that’s just the experimentalist in me talking now.

Good luck!

Jeff

That sounds about right. After I looked at the accelerometers on this site, I did some searching and found some engineering survey reports on the Internet regarding train vibration. They had some good information about how they took their measurements and what the ISO and British Standards were for vibration measurement. I want to make sure that my results are reasonably accurate according to standard measuring practice.

Seems that the high precision serial accelerometer shown here should be suitable to the job. And it fits my budget! I’ll order one soon and let you guys know how I get on. Thanks for the replies.

Ian

Ian,

a few weeks ago we’ve started working with the most recent SerAccel v3 accelerometer - the one you’ve just ordered. After developing some simple code within excel, which allows us to talk to our serial port we are able to record and post-process vibration data.

From our experience, you should not only report acceleration data, but also process for frequency, which is important, in case the train induced vibrations exite the fundamental mode of your building.

Best regards, let me know if you need some help,

Gregor, with http://www.lzatechnology.com/

Hey Gregor,

Any pros and cons to report with the SerAccel v3?

We are always interested to hear how customer use our products and what they think of the final product. Train noise/vibration sensing - never would have thought…

-Nathan

gregor:
From our experience, you should not only report acceleration data, but also process for frequency, which is important, in case the train induced vibrations exite the fundamental mode of your building.

Thanks for the advice. I understand from other sources that I should be looking at frequencies between 1Hz and 100Hz. But, I am not sure how I can measure more than one at a time without multiple sensors each set to different frequencies. Anyone have a thought on this?

Ok, now you’re looking into frequency domain stuff. I’m going to assume you have little understanding of frequency domain, so anything I (enclose in parentheses) from here on out, I’d suggest you Google.

To read 100Hz or below, you need to capture data at least at 200Hz (Nyquist Freqency). Now, greater sampling rates would be nice because you could use a digital filter (digital low pass filter) in your software to remove the higher frequencies and most of their adverse effects on your lower frequency measurements. From here on out, you’d need to run them through a spectral analysis algorithm (Fast fourier transform) to separate out the frequencies of the vibration. After you determine which frequencies are part of the train, you can track whenever those come through, thereby removing effects such as wind sway from the recorded accelerations.

Now, anyone have any good links for him? I’m still working on this stuff myself, except in a bit more depth :slight_smile: Otherwise, I’ll add on if I find something.

Here is a link to a paper on sampling, hope its not too late

http://www.johnsheatingandair.com/johnj … cture2.doc

Has anyone tried measuring rotating equipment vibration (i.e., from an engine) by use of the ACL-SerAccel-v3?

If one could get shaft position info (from a hall-effect sensor) and tack that onto the message containing the vibration info then we could tell where to add counterbalance weight to minimize the vibration.

Also by knowing speed we could derive IPS which is a pretty standard measure of rotating equipment vibration.

Is anyone interested in something like this or could someone provide some insight?

(I’ve worked extensively with the AVR processors but never with the PIC. Does the newer 20Mhz PIC used in the ACL-SerAccel-v3 have quite a bit more memory to support this additional code?)

72% space filled with SerAccel firmware v3.2 running on 16F88. A bit of room to play with.

-Nathan

Roger,

Measuring machinery vibration is kind of limited with this type of sensor. Vibration peaks occur at rotation rate and harmonics (i.e., 60 Hz for a 3600 RPM machine). Large peaks can generally be found at “forcing frequencies”, one of which would be the number of blades on a pump impeller x its rotation rate. The limited frequency range of these accelerometers is quickly exceeded and indeed, the accelerometer’s resonant frequency may be excited, leading to operload. Also, healthy machines can operate at fairly high acceleration levels. Accelerometers designed for machinery monitoring will usually handle 50 g vibration levels, but in some cases, even this isn’t a large enough range.

Engines and reciprocating machines in general are a tough case to handle with vibration analysis. The reciprocating motion and the explosions occuring each revolution generate a lot of harmonics and high frequency noise. Detecting meaningful changes from all the “grass” in the vibration signature is very difficult.

On the positive side, balancing rotation machinery with a sensor like this can be pretty simple. There is a method that doesn’t require any position (phase) reference, but it does require making 3 measurements with a trial weight located at different positions on the rotor. To see how this is done, please look here:

http://www.spintelligentlabs.com/SL-docs/an-6.pdf

There’s also quite a bit of background infomation about machinery vibration analysis on the web site too. It’s a well-proven technology that typically saves 3 to 11 x the cost of operating a monitoring program by finding small problems before they result in machinery failure.

I have a similar need for vibration analysis, but for helicopter rotor balancing. The rotors are tail rotor (2500 rpm) and main 2-blade rotor (450 rpm) and might typically see about 1g and 3 g readings for maximum vibrations. The last tail rotor I balanced was 0.33 ips at the start.

My question, is how do I go about setting up a balancer using an ADXL and a micro? I know I need an angular reference from a photo tach and reflective tape arrrangement or similar, and an accelerometer. Here is what I figure might work, am I wrong? Sample at the start of each angular zero crossing point, a full batch of accelerometer readings, to find the one in a full 360 degrees with the highest g value, then using counter readings, determine the angular position of the high reading. Am I oversimplifying this? Or do I somehow need an FFT?

Thanks in advance, it’s been difficult to locate any reference material for this subject, and the responses so far from others have been vague at best!

Forgive me if this seems stupid, but wouldn’t it be possible to just use the train tracks as an inductor, and transform the vibratory motion to an analog signal, then just use an ADC to measure frequency and magnitude? I can imagine that would be a little cheaper than an accelorometer, and if it was done right, I can’t imagine it would be much less accurate.

Alex

CORNWALLAV8R:
I have a similar need for vibration analysis, but for helicopter rotor balancing. The rotors are tail rotor (2500 rpm) and main 2-blade rotor (450 rpm) and might typically see about 1g and 3 g readings for maximum vibrations. The last tail rotor I balanced was 0.33 ips at the start.

My question, is how do I go about setting up a balancer using an ADXL and a micro? I know I need an angular reference from a photo tach and reflective tape arrrangement or similar, and an accelerometer. Here is what I figure might work, am I wrong? Sample at the start of each angular zero crossing point, a full batch of accelerometer readings, to find the one in a full 360 degrees with the highest g value, then using counter readings, determine the angular position of the high reading. Am I oversimplifying this? Or do I somehow need an FFT?

Thanks in advance, it’s been difficult to locate any reference material for this subject, and the responses so far from others have been vague at best!

The method you described is correct, it’s a brute force method. You have to sample at a very high rate to get peak value, otherwise you need a complex interpolation algorithm. But memory is cheap. Usually the sampling is done continously and a marker is provided that tells you where the once per rev signal is. You have to average 25 to 50 estimates before you can come up with any useful number.