I wonder if you guys can help. I am working on a project that consists of 4 sensors attached to a vehicle's gearbox those sensor measure vibrations at five gear shits at different loads for a certain period of time every once and while.
The sensors output data that is easily readable in excel format with a few tweaks so i have hundreds of .dat files from each sensor.
what i want to do is make something like a unique fingerprint/id for each shift and load. and when something out of the ordinary occure i get a message or an Alert.
What is the easiest way to achieve this? i am more of a mechanic than a software engineer.
I think you should time average your data in some way. Just from what I understand a gear shift occurs over ~1s frame so that will cut out some noise. May be easier to tell from there.
Also Python would be your best bet for this. Lookup "pandas" for python
Bentley Rogers
Thank you gonna look up pandas.
You are right about averaging the time this would be easy to achieve through excel for my intention is to fingerprint recorded data.
What i am afraid of is the programming procedure.
There is no software that outright does this right?
Landon Morris
>There is no software that outright does this right? I very much doubt it.
>What i am afraid of is the programming procedure. Collect detailed information on the task you want to get done, and on the result you want to have. Write shit down; on some paper if you're a man, but anything works. Use your search engine of choice to get info on how these little chunks work. Assemble the bits.
Use a scripting language like Python, Ruby or JavaScript (with NodeJS), avoid C/C++ if you don't care about speed and resource efficiency.
Good luck user, just come to /dpt/ if you end up getting stuck, maybe you'll even get help.
Jaxson Peterson
You could run some statistics functions on those data points and if there's a statistically significant change then that's what you're looking for. Since you're the mechanic you should first find out what exactly you are looking for, and then find out how to program it. Look up statistics first though, that would be my guess.
Ethan Taylor
Thank you user. [spoiler]btw /dpt/? whats that?[/spoiler] You rephrased my question anyway thanks.
Brandon Powell
Daily Programming Thread. It's another thread on Jow Forums.
Cooper Green
Fuck me everyone gonna think i am a newfag now for using spoilers. great. I swear i was lurking the whole time. I see. Avoided those threads like the plague for a long time.
>Fuck me everyone gonna think i am a newfag now >Caring about what other people think about you
>Avoided those threads like the plague for a long time. Not a bad idea. Sometimes there's interesting conversation to be found though.
Carter Wilson
>when something out of the ordinary occure i get a message or an Alert. define 'out of the ordinary'.
Benjamin Brooks
Anything out of the fingerprint behavior. For example the third shift has its own unique fingerprint if i insert a damaged gear or damage occure to a certain gear there would be a difference in the fingerprint and thus sending an alert message. it's not practical but its basic and my target for now. Only problem is i have a shitton of data but don't know how to farm them.
Aaron Flores
By fingerprint you mean the same pattern of vibrations happens over and over again, but when you insert a broken gear it's very different?
Luke Martinez
Yeah pretty much. Every shift combined with every load has a unique pattern/id the plot in the OP for example is a relation between time and magnitude of vibrations.
Aiden Johnson
If the measurement always looks kinda the same, you can probably use correlation (as used in dsp/digital signal processing). Correlation can detect a signal that you already know (good behaviour) in a sample (current behaviour), and the determines how similar they are. So if you always get the same pattern, there will be a high correlation between the measured and the good data, but if it's very different (broken gear) the correlation will be very low. However, this only works if you have an exact signal you're looking for, not if the data you measure is noise. You can use it in a radar system, where you send an exact signal and then listen for it to come back, but full of noise. You can still detect the signal in your measurement even though it's full of noise, because the correlation will be very high in one spot. If it's not like that, then it's probably not applicable
Robert Diaz
Yes it does look the same for the same shift the 4 sensors out put a similar plot. But for higher shifts the plots change drastically. So far i had no luck with Python panda as it is just an xcel replacement it seems. How could i apply the correlation method you are talking about?
Ian Russell
I'm just a student, one of my courses is DSP and we use matlab for it. mathworks.com/help/signal/ref/xcorr.html Matlab has it built in as a function, but to be honest I'm not sure how to code it manually, because I didn't study for the course yet.
This guy wants to prove the similarity between 2 audio files, and audio files are also just table data. And I assume you'll want to measure the similarity between your measured sensor data and a "good" sensor data. Then when the similarity is high, it's a good gear, if it's low, it's a bad gear.
But maybe you don't need all that and you can just take the average of all your data points or something, or look for spikes, I dunno
Xavier Murphy
Damn son that's really cool. You could try and fall for the machine learning meme, that'd probably be the easiest way so long as you can collect very large amounts of data.
Nicholas Cook
Beautiful i was looking for just a word to descripe my problem to search for and it turned out to be Correlation like you said not only that but the link you provided about the two audio files is fucking amazing as it is very similar to my case. Goddamn user i could kiss you. Now i have to wade through mat-lab which is disappointing but at least now i know where to start. Bless you mate.
Thank you user it's not a break through but i am getting there. I Can produce shitton of data in milliseconds. Gonna lookup machine learning too thanks.
Evan Howard
Good luck, user. Make sure correlation is actually applicable to your data. If you use matlab, you can get the correlation between 2 signals in like 3 lines of code since the functions are already there and there's probably tutorials about it on google. Also, it is possible to compile the matlab script so that you can use the compiled program together with another programming language. I haven't done this, but maybe you'd like to know about it, then you don't have to do everything in matlab, only the correlation.
Bentley Ross
can't you color code each excel column? they could act as your unique ID