1
$\begingroup$

I’m looking at some measurements for elevator door vibrations from this dataset (a cleaned version of what is available here, as I understand it) and, if I plot the series of measurements, I get the following image:

enter image description here

Measurements of some sensor are taken at 4Hz and it seems that, whatever the sensor is measuring, parts of the bottom signal/series are somehow shifted in time and amplitude. In addition, there seems to be some additional signals (the sparser ones) that are “mixed” with the rest.

My knowledge on this matter is close to zero, as you can guess by my almost complete lack of control over the terminology, but when I look at this some questions come to mind.

Is it actually possible to have measurements that look like this? Is it possible that they actually come from measurement series taken from different sensors and later superimposed? If so, what would one do to separate them? Or should one believe that someone manipulated the dataset adding some spurious data?

Thanks!

$\endgroup$

1 Answer 1

1
$\begingroup$

The data file appears to be partially corrupted. In the third column regularly appears a month name abbreviations (in German) instead of proper digits

93.636;73.99;18.028
93.632;73.99;18.029
93.628;73.989;18. Mrz
93.624;73.989;18.031
93.621;73.989;18.032

You need need to replace "Mrz" (Maerz or March) with "030", "Apr" with 040 etc.

There is also more weirdness towards the end of the file. Values starting with "12." have been replaced with "Dez "

63.344;73.962;12.783
63.343;73.962;12.781
63.343;73.961;Dez 78
63.343;73.961;12.779
63.342;73.961;12.777
63.342;73.961;12.776
63.342;73.961;12.775
63.341;73.961;12.774
63.341;73.961;12.772
63.341;73.961;12.771
63.34;73.961;Dez 77
63.34;73.96;12.769

This can all be repaired, but it'll take a little work.

Is it actually possible to have measurements that look like this?

Not really. That doesn't represent anything physical

Is it possible that they actually come from measurement series taken from different sensors and later superimposed?

Unlikely but you have to ask the source.

Or should one believe that someone manipulated the dataset adding some spurious data?

One should just take a look at the raw data file instead of blindly guessing. Just open it in Excel and take a look. If there are letters where you expect digits something is wrong.

This appears to be a simple transcription error. Apparently someone ran the data some sort of a number to date converter when saving.

$\endgroup$
1
  • $\begingroup$ Thanks for the answer @Hilmar. I should have probably specified that I ran some quantitative analyses on the version of that dataset which was published on kaggle and it threw no errors, it may have been cleaned. Still, I gather you believe that weird time series I'm seeing may be due to general dataset quality issues, right? $\endgroup$
    – imergodic
    Apr 28, 2022 at 6:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.