Intelligent symbol attributes extraction from engineering drawings
Helle, Riku (2023-06-29)
Intelligent symbol attributes extraction from engineering drawings
(29.06.2023)
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
avoin
Julkaisun pysyvä osoite on:
https://urn.fi/URN:NBN:fi-fe2023072691658
https://urn.fi/URN:NBN:fi-fe2023072691658
Tiivistelmä
Engineering drawings have a lot of information in them that can currently only be extracted by going through them manually. The goal for this thesis was to create a way to automatically extract symbol data from engineering drawings and use the extracted data to make the first demo application. The first application would calculate a complexity value for each drawing based on the extracted data.
The technologies selected for this task included text recognition and computer vision model YOLOv7 that was selected after testing different models. Machine learning based computer vision models were trained with different sized labelled training datasets and tested with a separate set of drawings meant for testing. The total amount of drawings used was 443.
The results achieved in the experiments reached good accuracy. The symbol with the most occurrences was surface roughness with over 5000 occurrences reaching an accuracy of 96\%. But due to problems caused by resolution, text recognition could not be fully implemented. Also, the complexity calculation was not further explored due to more work needed. So, future developments are needed, but the results show promising signs.
The technologies selected for this task included text recognition and computer vision model YOLOv7 that was selected after testing different models. Machine learning based computer vision models were trained with different sized labelled training datasets and tested with a separate set of drawings meant for testing. The total amount of drawings used was 443.
The results achieved in the experiments reached good accuracy. The symbol with the most occurrences was surface roughness with over 5000 occurrences reaching an accuracy of 96\%. But due to problems caused by resolution, text recognition could not be fully implemented. Also, the complexity calculation was not further explored due to more work needed. So, future developments are needed, but the results show promising signs.