During laser welding, a production line generates signals indicating various process conditions, such as plasma levels, laser intensity, and temperatures over time. The challenge is translating these signals into an accurate assessment of weld quality. This patented method combines convolutional neural networks with a graph neural network to not only analyze individual features but also their relationships. Samples collected during welding are categorized into quality groups and divided into base and test datasets. A convolutional network first extracts features for each category, transforming them into node features, with relationships represented as edges in a fully connected graph. The graph neural network then updates both node and edge features, incorporating relationship information into the model. Test features are processed through this trained model to evaluate weld quality in real time, enabling detection of issues during production rather than after defects occur. Within the broader MUST portfolio, this innovation exemplifies a shift in data representation, moving from analyzing features independently to understanding how they relate. Unlike conventional models that treat measurements as isolated feature vectors, graph learning explores inter-sample relationships. In complex physical systems, these relationships can carry vital information, and explicitly modeling them can enhance the interpretation of industrial sensor data, leading to more meaningful insights into manufacturing quality.
Turning welding signals into relationships AI can understand
Patent number: CN 112967231 B
Inventor(s): F. Deng, Y. Huang, G. Yao, H. Feng, W. Li, Y. Hu, H. Wang, Y. Ding, D. Zhong, J. Xi, and N. Li
Citation: F. Deng, Y. Huang, G. Yao, H. Feng, W. Li, Y. Hu, H. Wang, Y. Ding, D. Zhong, J. Xi, and N. Li, “焊接质量检测方法及其装置、计算机可读存储介质 [Welding-quality detection method and apparatus, and computer-readable storage medium],” China Patent CN 112967231 B, Nov. 15, 2022.
