An additive manufacturing online defect monitoring method based on semi-supervised learning
By employing a semi-supervised learning method that combines dual-view sample selection and a ternary hybrid loss function, the problem of low accuracy and poor robustness caused by noise labels in additive manufacturing is solved. This enables efficient and robust online quality monitoring, adaptable to various additive manufacturing processes.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
- Filing Date
- 2026-06-12
- Publication Date
- 2026-07-17
AI Technical Summary
In existing online quality monitoring methods for additive manufacturing, noise labels lead to low model prediction accuracy and poor robustness, making it difficult to adapt to the complex data distribution in industrial scenarios and affecting the large-scale application of deep learning methods.
A semi-supervised learning-based approach is adopted, which uses a dual-view sample selection and a ternary mixture loss function to filter clean and noisy samples. Supervised training of clean samples and unsupervised training of noisy samples are used, combined with data augmentation and cosine similarity regularization, to achieve efficient training of the model.
It improves the model's prediction accuracy and robustness under noisy labeling conditions, enhances the model's adaptability to industrial scenarios, reduces hardware costs, and meets the needs of real-time quality monitoring.
Smart Images
Figure CN122415596A_ABST