Title:	Sim to Real: Automatic manufacturing quality inspection based on computer vision
DNr:	Berzelius-2023-368
Project Type:	LiU Berzelius
Principal Investigator:	Atsuto Maki <atsuto@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2024-01-30 – 2024-08-01
Classification:	10207
Keywords:

Abstract

A challenge to applying deep learning-based computer vision technologies for manufacturing quality inspection lies in the cost, time, and manual effect of collecting a large amount of annotated training data. This project aims to solve this challenge by training a model with only synthetics data generated from manufacturing CAD models. The method should be able to ignore the gap between the simulation and real data, which means the model will only be trained with synthetics data, but be tested on real data and achieve promising results. In the previous Berzelius project, we tested different deep learning models (including ViT-based models like DINO and CNN-based models like Yolov8 ) on synthetic data. In this project, we will focus on the synthetic data-generating pipeline. We want to dynamically improve model performance by iteratively generating synthetic data, training models, and then using the insights gained to enhance the next data generation cycle until optimal performance is achieved. This project is related to Scania CV AB and KTH, and partially supported by the Wallenberg AI, Autonomous Systems and Software Program (WASP) funded by the Marianne and Marcus Wallenberg Foundation. Besides, Jacob from Uppsala University will work with Scania on this project in the next six months.