Towards generalizable machine learning prediction of downskin surface roughness in laser powder bed fusion
Bibliographic record
Abstract
Downskin surface quality of laser powder bed fusion (L-PBF) remains a challenge due to the complex, multi-scale physics governing it. While numerical or experimental approaches alone can be significantly resource intensive, data-driven approaches such as machine learning (ML) have the potential to be more practical. However, the generalizability of ML models currently reported in literature is unclear; few ML models can predict reliably outside of their training domain. This study addresses these challenges by (i) demonstrating a downskin surface roughness classification model, trained on the largest reported dataset for downskin roughness ( ∼ 400 downskin specimens spanning five builds and two ferrous alloys) and (ii) conducting a thorough investigation of the model’s generalizability. Additionally, this study highlights critical issues such as data imbalance, generalization to unseen data, and the importance of rigorous evaluation. By implementing robust ML practices, we focused on model performance across different training and evaluation domains. Our findings indicate satisfactory performance when using the more conservative balanced accuracy metric, achieving about 95% inter-domain and 83% intra-domain accuracy. Although there is still room for improvement, these results demonstrate a significant reduction in the risk of overfitting, thereby enhancing the classifier’s generalizability. This work underscores the importance of methodological rigor in machine learning applications, advocating for greater attention to data treatment and evaluation strategies. This approach may ultimately lead to more effective and usable ML models. The data-centric results indicated that (i) physics-informed features can improve performance during domain shifts, and (ii) increased the size and variety of datasets allows even computationally light models to achieve favorable performance.
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 | 0.002 | 0.005 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| 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".