2 Replies - 197 Views - Last Post: 28 July 2019 - 09:17 AM Rate Topic: ***-- 2 Votes

#1 istemihan90   User is offline

  • New D.I.C Head

Reputation: 4
  • View blog
  • Posts: 4
  • Joined: 28-July 19

Deep learning for EMC tests

Posted 28 July 2019 - 05:58 AM

Hello to all!

I am an electronics engineer and working as a hardware design engineer (I usually design Mainboard for Televisions). However, I am not unfamiliar with programming software, I am particularly good at embedded C with microcontroller applications.

So I just started using python. I have a project that is about deep learning of EMC (Electro Magnetic Compability) traits in televisions. Let me explain it in more detail. Every electronic device in the market is obliged to meet universal EMC regulations. One (and probably the most important one) of the EMC regulation tests is REM (radiated emission measurements) which measures the electromagnetic radiation emitted from the device under test by each angle, height and some other conditions. After we design and produce the television, before releasing the TV on market, we put the final product in EMC tests and in 99% of the cases, there is a REM problem. Do overcome this REM problem we spend hours to days even weeks in EMC laboratories (which is quite expensive). We do some certain types of modifications in the TV to meet the regulations. I will give you an example below. This is the first REM test of a product I designed:
https://ibb.co/FnM7mJW
https://ibb.co/ZKGDCD8
You will see the red thick line that represents regulation limits starting from 40 dBuV/m. You will also notice that the radiation level around 220 MHz is critically close to the limits. To prevent this risk I switched one of the screws in the mainboard from metal to isolated plastic and the new result was as below:
https://ibb.co/6H1pHQF
https://ibb.co/Rz16Mwn
now it is good to go. However, the thing is I had to spend almost 2 hours to figure out what the source of the problem was. Because, your hunches are usually wrong when you work on EMC and there are like 1000 methods to try. But luckily, if you try that 1000 methods you definitely solve the problem. So I thought, why don't I just collect all the REM data (from 2005 to now there are 3000+ data like this), feed them in to a deep learning application and make the algorithm do the job for me in the most efficient way.

The plan is to develop a regression model for REM tests. The input data will be the size (40"/ 55"/ 70" etc), refreshing Rate, frame rate, cable types, panel vendor etc. all the categorizable inputs will be used for regression model. I will add all the data from 2005 and feed them into algorithm, and algorithm will learn which method resulted in which outputs (like decrease in frequency range 100 MHz by 10 db etc) in which input and will build a model. After the model is ready, it will make suggestions for failed EMC test results. I hope I explained my idea clearly.

I started by following through this tutorial: https://stackabuse.c...pl...cognition/

I can detect the optical characters in the png file. After that I'm planning to extract the data from the graph (I don't know where to start here). After that, I will start to learn the ways to build a regression model for this particular application.

What I would like to learn from you is a guideline for this project. Because I'm buzzled! I don't know where to start! There are many applications which seem useful to me and python is a really powerful script and has a great community.

Could you give me advice and write an outline - an action plan for me? Step by Step what should I learn, which topic I should study to achieve what? It would be the most helpful thing for me at this stage. Thanks to you all!

Is This A Good Question/Topic? 1
  • +

Replies To: Deep learning for EMC tests

#2 modi123_1   User is online

  • Suitor #2
  • member icon



Reputation: 15268
  • View blog
  • Posts: 61,212
  • Joined: 12-June 08

Re: Deep learning for EMC tests

Posted 28 July 2019 - 08:36 AM

Quote

I can detect the optical characters in the png file. After that I'm planning to extract the data from the graph (I don't know where to start here).

That seems like a bad way of going about it. Why not just get the data that created the graph and use that?

Also seen here:
https://python-forum...g-for-EMC-tests
Was This Post Helpful? 0
  • +
  • -

#3 istemihan90   User is offline

  • New D.I.C Head

Reputation: 4
  • View blog
  • Posts: 4
  • Joined: 28-July 19

Re: Deep learning for EMC tests

Posted 28 July 2019 - 09:17 AM

View Postmodi123_1, on 28 July 2019 - 08:36 AM, said:

Quote

I can detect the optical characters in the png file. After that I'm planning to extract the data from the graph (I don't know where to start here).

That seems like a bad way of going about it. Why not just get the data that created the graph and use that?

Also seen here:
https://python-forum...g-for-EMC-tests

Thanks for your response. The reason I need to extract the data from the graph is beacuse Ican not reach the raw data of the older test results. Also, the raw output of the system(which we cannot configure. It is a huge system of rhode schwarz and can not be programmed) takes too much time (like 25 minutes) and the excel file is too big. I think extracting the data from a simple graph is more convenient. I need it to get the older data anyway.

Also even if it was not the problem, I still need an action plan, libraries to work. I need yo find my way quickly.

right now I m trying to detect color on the mage map than detect the limit line than extract the data per 1 Mhz and write them in a csv file.
Was This Post Helpful? 1
  • +
  • -

Page 1 of 1