Leakage-Safe Machine Learning for Hydrogen Embrittlement Detection in 316L Stainless Steel: A Region-Held-Out Evaluation of Texture and Deep Features in SEM Micrographs

Muhammad Awais; Muhammad Yaseen; Abdul Shakoor; Niaz Ahmed Niaz; Huria Zia; Muhammad Zain Shakoor

cs.LGSep 23, 2026

Leakage-Safe Machine Learning for Hydrogen Embrittlement Detection in 316L Stainless Steel: A Region-Held-Out Evaluation of Texture and Deep Features in SEM Micrographs

Authors: Muhammad Awais, Muhammad Yaseen, Abdul Shakoor, Niaz Ahmed Niaz, Huria Zia, Muhammad Zain Shakoor

Organizations: Dipartimento di Fisica e Astronomia “G. Galilei”, Università di Padova, Italy · Institute of Physics, Bahauddin Zakariya University, Multan, Pakistan

Abstract

Scanning electron microscopy (SEM) is routinely used to characterize the microstructural changes caused by hydrogen embrittlement (HE) in structural steels. Machine learning can automate this characterization, but models are often evaluated using image-level splits. When several images come from the same specimen region, such splits leak information between the training and test sets. Here, we propose a region-held-out protocol for classifying as-received (AR) and hydrogen-charged (H2) SEM micrographs of 316L stainless steel, based on Leave-One-Region-Out (LORO) cross-validation over 14 spatial regions (8 AR, 6 H2; 31 images). We compared six feature-classifier combinations built on local binary patterns (LBP), grey-level co-occurrence matrices (GLCM), self-supervised convolutional embeddings pretrained on 143 unlabeled SEM images, and a convolutional neural network (CNN). The simplest texture approach, LBP with a support vector machine (LBP+SVM), performed best, achieving a balanced accuracy of 0.79, H2 recall of 0.69, and H2 precision of 0.82, outperforming every deep-learning and combined-feature model. A group-level permutation test (500 permutations sampled from the 3,003 possible region-to-label assignments) yielded p = 0.008, indicating that the result cannot be explained by a chance alignment of the region structure. Grad-CAM maps from a CNN trained on the full dataset tended to concentrate on localized surface and grain-boundary features, where hydrogen-induced morphological changes are known to occur. Under a leakage-safe, statistically validated protocol, texture descriptors recover a hydrogen-charging signature from SEM micrographs even with few samples, and the same protocol can be extended to larger HE detection studies in other alloy systems.

Figures & tables

Model	Balanced accuracy	F1 (H 2 )	Precision (H 2 )	Recall (H 2 )
LBP+SVM	$0.79\pm 0.36$	0.75	0.82	0.69
CNN	$0.67\pm 0.45$	0.67	0.55	0.85
LBP+GLCM+SVM	$0.62\pm 0.43$	0.52	0.60	0.46
LBP+GLCM+RF	$0.59\pm 0.45$	0.50	0.55	0.46
SSL_global	$0.56\pm 0.43$	0.48	0.50	0.46
GLCM+SVM	$0.52\pm 0.46$	0.36	0.44	0.31

Table 1: LORO cross-validation results (316L AR vs. H 2 , 14-fold).

	Predicted AR	Predicted H 2
Actual AR ( $n=18$ )	16	2
Actual H 2 ( $n=13$ )	4	9

Table 2: Pooled confusion matrix (LBP+SVM, 31 images).

Figure 1: Pooled confusion matrix for the LBP+SVM model under Leave-One-Region-Out cross-validation, aggregated across all 14 folds (31 images in total: 18 AR, 13 H 2 ). Diagonal cells are correct classifications; off-diagonal cells are misclassifications. The model reaches a specificity of 0.89 for AR and a recall of 0.69 for H 2 , with 4 false negatives (H 2 classified as AR) and 2 false positives (AR classified as H 2 ).

Figure 2: Mean balanced accuracy ( $\pm$ standard deviation) for each of the six model configurations under 14-fold Leave-One-Region-Out cross-validation. The chance level (BA $=0.50$ ) is shown as a reference line. LBP+SVM has the highest mean balanced accuracy but also a large variance, a consequence of the small per-fold sample sizes that come with region-level evaluation.

Figure 3: Per-fold balanced accuracy of the six models across the 14 held-out regions. Darker and lighter cells indicate higher and lower balanced accuracy in that region. Regions with consistently low accuracy across several models are candidates for closer inspection.

Figure 4: Null distribution of balanced accuracy for the LBP+SVM model, generated from 500 permutations of the region-to-label assignment under the Leave-One-Region-Out protocol. The observed balanced accuracy (0.79, dashed red line) sits in the far right tail of the null distribution ( $p=0.008$ ). The chance level (0.50) and the 95th percentile of the null distribution are shown for reference.

Figure 5: Grad-CAM activation maps overlaid on representative SEM images of three AR and three H 2 specimens, generated from a CNN trained on the full 316L dataset (20 epochs, no held-out split). Warmer colors mark regions with more influence on the model’s prediction.

Figure 6: Pixel-wise mean image computed across all H 2 -labeled 316L specimens, showing consistent large-scale intensity and texture patterns associated with the hydrogen-charged condition.

Figure 7: Region-level comparison of texture characteristics between the AR and H 2 specimen groups.

Figure 8: Distribution of Local Binary Pattern histogram values for H 2 -labeled images, showing the texture feature space used by the LBP+SVM classifier.

Explore similar work

Jun 8, 2026cond-mat.mtrl-sci

Context-Aware Deep Learning for Defect Classification in Atomic-Resolution STEM

Artificial intelligence is rapidly advancing materials characterization, yet most applications in electron microscopy rely solely on image contrast, overlooking the chemical and experimental context that shapes image formation. This limitation makes defect classification inherently ambiguous, as similar contrasts can arise from different materials or imaging conditions. Here we develop a context-aware learning framework that integrates image-derived contrast with metadata describing composition, beam energy, and detector geometry. Using a systematically constructed dataset of ~55 million simulated patches spanning 576 cases across 96 doped monolayer transition-metal dichalcogenides, we show that conditioning on contextual variables transforms defect classification from an ill-posed image-only task into a well-posed, physically grounded problem. The framework achieves over 98% accuracy on simulations and near-human agreement on experimental data, with a 94% reduction in posterior entropy. By emphasizing contextual grounding over architectural complexity, this approach links experimental image contrast to the underlying chemical and imaging conditions, supporting physically grounded defect assignments and a general pathway toward multimodal AI models for autonomous materials characterization.

Jiadong Dan, Cheng Zhang, Leyi Loh +5

Department of Biological Sciences, National University of Singapore, Singapore · Centre for BioImaging Sciences (CBIS), National University of Singapore, Singapore · Department of Materials Science and Engineering, City University of Hong Kong, China +5

Electron Microscopy Defect Detection Materials Science Adaptive Histogram Equalization Deep Learning Contextual Grounding

Ask PDF

Aug 22, 2025cs.CV

EM3M: An Electron Micrograph Dataset for Microstructural Segmentation and Generation

Quantitative microstructural characterization is fundamental to materials science, and electron micrographs (EMs) provide indispensable high-resolution insights. However, progress in deep learning-based analysis of EMs has been hampered by the scarcity of large-scale, expert-annotated public datasets. To address this issue, we introduce EM3M, a large-scale and multimodal dataset for instance-level understanding of EMs. EM3M comprises 5,091 high-quality EMs, approximately 3 million instance segmentation annotations, and image-level textual descriptions with disentangled attributes. The dataset is constructed through a rigorous multi-stage curation and validation pipeline, with comprehensive statistical analyses to ensure reliability and reproducibility. Building upon these curated image-text pairs, we further provide a text-to-image diffusion model that serves as a controllable data augmentation engine, demonstrating that synthetic augmentation consistently improves downstream segmentation performance. To establish a systematic benchmark, we evaluate representative instance segmentation methods on EM3M. Our results reveal that conventional detection-based and query-based methods struggle with the extreme instance densities and textural complexities inherent in EMs. We additionally provide an optimized flow-based baseline to facilitate fair comparison and future research. EM3M {Dataset: https://huggingface.co/datasets/UniParser/EM3M}, the generative engine {Generation: https://huggingface.co/UniParser/EM3M-Gen}, and an online demo {Segmentation demo: https://www.bohrium.com/apps/uni-aims} are publicly available to support future research in automated materials analysis.

Nan Wang, Zhiyi Xia, Yiming Li +7

DP Technology, Beijing, China

Electron Microscopy Porous Media Insights

Ask PDF

Jul 30, 2026cs.LG

Predicting Steel Fatigue Life from Micrographs Using Physics-Informed Deep Learning

Here is the plain text version optimized for arXiv's submission form. Custom macros (like \CV and \SI) have been converted to standard text/math so they render correctly on the webpage: Evaluating the fatigue life of structural steels conventionally requires mechanical testing lasting tens to hundreds of hours, making it impractical for rapid quality control. We present CV, a computer vision framework that estimates the fatigue life (

\log N_f

) of lightweight alloy steels directly from optical micrographs without physical testing.The pipeline features a seven-stage OpenCV preprocessing routine to remove artifacts, a 28-dimensional physics-informed feature extractor (quantifying crack morphology, grain structure, porosity, and texture), and a CNN regression model trained with a Gaussian negative log-likelihood (GNLL) loss to jointly predict

\log N_f

and sample-specific uncertainty

\hatσ

.Evaluating three architectures (SE-CNN, ResNet-50, VGG-16) on a synthetic micrograph benchmark, ResNet-50 achieves

R^2 = 0.93

, RMSE = 0.18 log-cycles, and macro-F1 = 0.91. The GNLL objective reduces Expected Calibration Error by 76% compared to a mean-squared-error baseline (ECE:

0.089 \rightarrow 0.021

). Grad-CAM maps confirm the network attends to metallurgically meaningful microstructural features.Running in under 65 ms per image, the pipeline and synthetic dataset generator are open-sourced. Because validation relies entirely on synthetic micrographs, these results demonstrate methodological soundness under simulated conditions; a domain-transfer study on real field samples is the immediate next step.

Aryuemaan Kumar Chowdhury

Department of Lightweighting Engineering, Indian Institute of Technology Hyderabad, India · Research & Development, Oscowl AI, Hyderabad, India

Shape-Memory Alloy Porous Media Useful Life Prediction Computer Vision Deep Learning

Ask PDF