So, you are out on a job and there is a noisy transformer – you know, that humming noise that we are all familiar with. However, something doesn’t sound quite right. A discussion begins about whether the transformer is seeing harmonics, overloading, or some other condition. It comes down to opinion until you can actually see something. You don’t have ultrasound or other sound analysis systems nearby, so what can you do in a pinch.
Along comes Python and Jupyter Notebooks to the rescue. For this article, the Jupyter Notebook and associated sound file can be found at MotorDoc1/FunWithSound (github.com). And, no, you don’t have to use it for just a transformer, although you will want some Python experience.
Yes, yes, there are apps that do this, but what fun is that to a data scientist? Or just about any engineer?
IEEE Standard C57.136-2000, “IEEE Guide for Sound Level Abatement and Determination for Liquid-Immersed Power Transformers and Shunt Reactors” identifies specific sound frequencies and what they relate to. This can provide some direction while you are working on obtaining more advanced systems to verify the results.
Here is the sound:
Now, when processed through the Python code (and some modifications made) we end up with spectra in which we have amplitude versus frequency. In Figure 1 we have relative log values and in Figure 2 we have a spectra in -dB where we’ve ‘zero-d’ the highest peak. We will discuss how to use your phone as a dB meter in a separate article.
Here is what we found with the above sound which is at 44,000 samples and across 5 seconds. Note that sound is position and direction-dependent.
As you can see, data of this type is useful in setting investigative direction. The fun part about doing this in Python (or MatLab) is that you can manipulate the data in order to filter, etc. beyond what is done in most phone apps.
More on this in the next article.
Note: this article is for general information and implementation/use does not imply warranty.
