Related Experiment Video
Updated: Oct 30, 2025
11:34
Subsurface Defect Localization by Structured Heating Using Laser Projected Photothermal Thermography
Published on: May 15, 2017
11.3K
A Spatio-Temporal Ensemble Deep Learning Architecture for Real-Time Defect Detection during Laser Welding on Low
Christian Knaak1, Jakob von Eßen1, Moritz Kröger2
1Fraunhofer-Institute for Laser Technology ILT, Steinbachstrasse 15, 52074 Aachen, Germany.
Sensors (Basel, Switzerland)
|July 2, 2021
Summary
This study introduces an ensemble deep learning model for real-time welding defect detection. The system accurately identifies critical weld flaws using infrared imaging, enhancing manufacturing quality control.
Area of Science:
- Manufacturing Engineering
- Artificial Intelligence
- Materials Science
Background:
- Advanced process monitoring is crucial for diagnosing production issues and ensuring final component quality.
- Real-time product quality recognition is essential for developing autonomous and self-improving manufacturing systems.
Purpose of the Study:
- To investigate a novel ensemble deep learning architecture for intelligent process monitoring and weld defect detection.
- To extract and utilize spatio-temporal features from infrared image sequences for identifying critical welding defects.
Main Methods:
- Developed a novel ensemble deep learning architecture combining convolutional neural networks (CNN), gated recurrent units (GRU), k-nearest neighbors (kNN), and support vector machines (SVM).
- Extracted spatio-temporal features from infrared image sequences, focusing on keyhole and weld pool regions.
- Evaluated the architecture against classical machine learning and state-of-the-art deep learning methods using a comprehensive validation scheme and grid search for hyperparameter optimization.
Main Results:
- The proposed ensemble deep learning method achieved the highest detection rates and most robust weld defect recognition compared to all investigated methods.
- The system effectively located critical welding defects such as lack of fusion, sagging, lack of penetration, and geometric deviations.
- The optimized ensemble deep neural network demonstrated low latency (1.1 ms) on embedded computing devices, suitable for real-time applications.
Conclusions:
- The novel ensemble deep learning architecture provides an effective solution for real-time, intelligent process monitoring in manufacturing.
- The method significantly improves the accuracy and robustness of weld defect recognition, contributing to enhanced quality control.
- The system's optimization for embedded devices enables practical implementation in real-time industrial applications.
