Investigating the Performance of a Vision Transformer Model for Anomaly Detection in Laser Metal Deposition Imaging
Bibliographic record
Abstract
Laser metal deposition (LMD) is recognized as a critical technique in Additive Manufacturing (AM) that allows the production and repair of components in a high-quality, efficient, and cost-effective manner. However, defects may still arise in the deposited components. While conventional architectures like Convolutional Neural Networks (CNNs) have shown satisfactory results in detecting these defects using images captured during the process, transformer-based models remain relatively underexplored in this context. This study focused on designing a transformer-based architecture that could achieve high accuracy in identifying anomalies through melt pool images obtained during the wirefed LMD process. Upon its development, it was used to crossreference the predictions of an existing powerful CNN approach to ensure the reliability of its outcomes. Initially, the algorithm was trained using a custom Vision Transformer-decoder architecture with no labels involved, resulting in an accuracy of 92.66%. By utilizing the captured information from the classification token, its ability to identify anomalies was significantly improved, achieving 99.78% in a 900-image dataset. However, when evaluated on 6,497 unseen frames from the process with ground truth predictions generated by the CNN model, ViT’s accuracy decreased to 97.83%, a result attributed to the specific training method and the variability in the test set. Despite this reduction, the results were considered satisfactory, given the relatively new application of transformers on images, which has not been extensively explored in the field of anomaly detection. Overall, this research offers a comprehensive explanation of the proposed model architecture and outlines the necessary modifications required to achieve near-perfect performance on a transformer-based architecture, paving the way for future enhancements in anomaly detection.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".