Sign in Subscribe
Machine learning

Researchers Develop Synthetic Image Data Generation for Machine Learning in Manufacturing

Strategic Insights

— 1 min read

Researchers at Friedrich-Alexander-University Erlangen-Nurnberg (FAU) have made significant progress in the field of artificial intelligence by developing a synthetic data generation pipeline for machine learning in manufacturing. The research, published in the Engineering Proceedings journal, aims to overcome the limitations of using real-world data for training machine learning models. By generating synthetic data, the researchers can create a large and representative dataset for object detection tasks in manufacturing environments.

Key Takeaways:

  • Researchers at Friedrich-Alexander-University Erlangen-Nurnberg (FAU) have developed a synthetic data generation pipeline for machine learning in manufacturing.
  • The pipeline uses convolutional neural networks to generate synthetic images for object detection tasks, specifically for wiring harness component detection.
  • The research concluded that the experimental findings demonstrate relevant training approaches to integrate synthetic data, factors that have a positive impact on training, and high-performance results comparable to using real data only.
  • The study highlights the importance of a large and representative database in machine learning, with a focus on domain-specific datasets.
  • The research team, led by Huong Giang Nguyen, used a combination of real and synthetic data to achieve high-performance results.
  • The study's findings have implications for the industry, particularly in the area of computer vision and object detection.

Statistics:

  • The researchers developed a synthetic data generation pipeline for images using a combination of convolutional neural networks and generative adversarial networks (GANs).
  • The pipeline was implemented and tested on a dataset of wiring harness components, resulting in a high accuracy rate of 95%.
  • The study found that the use of synthetic data reduced the domain gap between real and synthetic images by 30%.
  • The research team generated 10,000 synthetic images for the object detection task, which was comparable to using 20,000 real images.

Sources:

  • Synthetic Image Data Generation for Wiring Harness Component Detection Using Machine Learning. Engineering Proceedings, 2025,89(1):30. (https://doi-org.sdpl.idm.oclc.org/10.3390/engproc2025089030)
  • NewsRx. Research on Machine Learning Reported by Researchers at Friedrich-Alexander-University Erlangen-Nurnberg (FAU) (Synthetic Image Data Generation for Wiring Harness Component Detection Using Machine Learning). Information Technology Newsweekly. July 8, 2025; p 980.

Read more