Martin Müller
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Martin Müller has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- Frank Mücklich
Department of Materials Science, Saarland University, 66123 Saarbrücken, Germany
3 shared publications - Björn-Ivo Bachmann
Department of Materials Science, Saarland University, 66123 Saarbrücken, Germany
2 shared publications - Marie Stiefel
Department of Materials Science, Saarland University, 66123 Saarbrücken, Germany
2 shared publications Andrew Kitahara
Analytical Mechanics Associates, Hampton, USA
1 shared publication- Camilo Andres Martinez Martinez
Department for Material Science, Saarland University, Saarbrücken, Germany
1 shared publication D. Britz
Materials Engineering Center Saarland (MECS) , Saarbrücken , Germany
1 shared publication
Latest Publications
- Open Access
- 25 September 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1) , 29853
A systematically designed open-access dataset for cross-microscope machine learning benchmarking in complex steel microstructure classification
The establishment of robust machine learning workflows for microstructure analysis requires datasets that systematically capture the variance introduced by the imaging pipeline. While existing studies in this field predominantly rely on well-curated, homogeneous datasets optimized for high classification performance, these conditions rarely reflect the variance encountered in real-world settings. This work presents a publicly available, FAIR-compliant dataset of 5327 annotated micrograph patches, showing bainitic and martensitic microstructural constituents in quenched and quenched-and-tempered low-carbon steels, acquired on three different optical microscopes across two magnification levels (20x and 50x) and under deliberately varied acquisition conditions. Identical regions of interest were imaged on all platforms and aligned by image registration, enabling optimal comparability by pixel-wise alignment across instruments and magnifications. This correlative structure makes the dataset uniquely suited for systematic investigations of cross-instrument and cross-magnification domain generalization. Using this dataset, the influence of microscopy modality-induced domain shifts on the classification performance of deep convolutional neural networks was analyzed. A general model trained on the full dataset achieved near-perfect classification (F1 score of 99.91%), demonstrating that modern deep learning is capable of handling the full range of imaging-induced variance when sufficient data coverage is provided. Cross-domain experiments revealed that generalization is governed primarily by the physical fidelity of the input representation and the coverage of imaging-induced variance in the training data, rather than by model architecture or nominal magnification alone. The direction in which resolution gaps are bridged between training and validation domains proved more decisive than information density per se. Among single-instrument configurations, broad variation of acquisition conditions, such as aperture, exposure, and color temperature, contributed more effectively to cross-domain robustness than optical diversity between objective lenses alone. These results suggest that targeted variation of acquisition conditions, rather than purely morphological diversity, is a particularly effective strategy for improving model generalization. The dataset is released to the community as a benchmark resource for domain shift analysis, generative modeling, and domain-aware learning in microstructure-based machine learning.
- 25 September 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1) , 29853
Open AccessA systematically designed open-access dataset for cross-microscope machine learning benchmarking in complex steel microstructure classification
The establishment of robust machine learning workflows for microstructure analysis requires datasets that systematically capture the variance introduced by the imaging pipeline. While existing studies in this field predominantly rely on well-curated, homogeneous datasets optimized for high classification performance, these conditions rarely reflect the variance encountered in real-world settings. This work presents a publicly available, FAIR-compliant dataset of 5327 annotated micrograph patches, showing bainitic and martensitic microstructural constituents in quenched and quenched-and-tempered low-carbon steels, acquired on three different optical microscopes across two magnification levels (20x and 50x) and under deliberately varied acquisition conditions. Identical regions of interest were imaged on all platforms and aligned by image registration, enabling optimal comparability by pixel-wise alignment across instruments and magnifications. This correlative structure makes the dataset uniquely suited for systematic investigations of cross-instrument and cross-magnification domain generalization. Using this dataset, the influence of microscopy modality-induced domain shifts on the classification performance of deep convolutional neural networks was analyzed. A general model trained on the full dataset achieved near-perfect classification (F1 score of 99.91%), demonstrating that modern deep learning is capable of handling the full range of imaging-induced variance when sufficient data coverage is provided. Cross-domain experiments revealed that generalization is governed primarily by the physical fidelity of the input representation and the coverage of imaging-induced variance in the training data, rather than by model architecture or nominal magnification alone. The direction in which resolution gaps are bridged between training and validation domains proved more decisive than information density per se. Among single-instrument configurations, broad variation of acquisition conditions, such as aperture, exposure, and color temperature, contributed more effectively to cross-domain robustness than optical diversity between objective lenses alone. These results suggest that targeted variation of acquisition conditions, rather than purely morphological diversity, is a particularly effective strategy for improving model generalization. The dataset is released to the community as a benchmark resource for domain shift analysis, generative modeling, and domain-aware learning in microstructure-based machine learning.
- Open Access
- 17 August 2026
- Published by Springer Nature in Metallography, Microstructure, and Analysis
- Vol. 15 (4) , 834-835
Special Issue Preview: Artificial Intelligence in Metallography
- 17 August 2026
- Published by Springer Nature in Metallography, Microstructure, and Analysis
- Vol. 15 (4) , 834-835
Open AccessSpecial Issue Preview: Artificial Intelligence in Metallography
- 3 August 2026
- Published by Springer Nature in Metallography, Microstructure, and Analysis
- Vol. 15 (4) , 836-847
Advancing AI-Driven Microstructure Analysis Through Correlative Microscopy Approaches
- 3 August 2026
- Published by Springer Nature in Metallography, Microstructure, and Analysis
- Vol. 15 (4) , 836-847
Advancing AI-Driven Microstructure Analysis Through Correlative Microscopy Approaches
- Open Access
Synthetic Rebalancing of Imbalanced Macro Etch Testing Data for Deep Learning Image Classification
The adoption of artificial intelligence (AI) in industrial manufacturing lags behind research progress, partly due to smaller, imbalanced datasets derived from real processes. In non-destructive aerospace testing, this challenge is amplified by the low defect rates of high-quality manufacturing. This study evaluates the use of synthetic data, generated via multiresolution stochastic texture synthesis, to mitigate class imbalance in material defect classification for the superalloy Inconel 718. Multiple datasets with increasing imbalance were sampled, and an image classification model was tested under three conditions: native data, data augmentation, and synthetic data inclusion. Additionally, round robin tests with experts assessed the realism and quality of synthetic samples. Results show that synthetic data significantly improved model performance on highly imbalanced datasets. Expert evaluations provided insights into identifiable artificial properties and class-specific accuracy. Finally, a quality assessment model was implemented to filter low-quality synthetic samples, further boosting classification performance to near the balanced reference level. These findings demonstrate that synthetic data generation, combined with quality control, is an effective strategy for addressing class imbalance in industrial AI applications.
Open AccessSynthetic Rebalancing of Imbalanced Macro Etch Testing Data for Deep Learning Image Classification
The adoption of artificial intelligence (AI) in industrial manufacturing lags behind research progress, partly due to smaller, imbalanced datasets derived from real processes. In non-destructive aerospace testing, this challenge is amplified by the low defect rates of high-quality manufacturing. This study evaluates the use of synthetic data, generated via multiresolution stochastic texture synthesis, to mitigate class imbalance in material defect classification for the superalloy Inconel 718. Multiple datasets with increasing imbalance were sampled, and an image classification model was tested under three conditions: native data, data augmentation, and synthetic data inclusion. Additionally, round robin tests with experts assessed the realism and quality of synthetic samples. Results show that synthetic data significantly improved model performance on highly imbalanced datasets. Expert evaluations provided insights into identifiable artificial properties and class-specific accuracy. Finally, a quality assessment model was implemented to filter low-quality synthetic samples, further boosting classification performance to near the balanced reference level. These findings demonstrate that synthetic data generation, combined with quality control, is an effective strategy for addressing class imbalance in industrial AI applications.
- 1 September 2025
- Published by Walter de Gruyter GmbH in Practical Metallography
- Vol. 62 (9-10) , 714-731
Deep learning-based precipitate quantification in STEM images of complex steel microstructures
Understanding precipitate evolution in microalloyed high-strength low-alloyed (HSLA) steels is essential for optimizing their processing and mechanical properties. In this work, we present an automated workflow for precipitate quantification in scanning transmission electron microscopy (STEM) images of niobium and titanium HSLA steels, combining deep learning-based semantic segmentation with conventional image processing techniques. Among the different STEM image contrasts tested, the high-angle annular dark-field-based model delivered the best segmentation performance, achieving a mean intersection over union of 0.8111 and a deviation of only 5.37 % in the measured mean particle diameter. The workflow’s robustness and generalizability were demonstrated on unseen images of different HSLA steels, with particle counts and size distributions aligning well with expectations based on the respective alloying concepts.
- 1 September 2025
- Published by Walter de Gruyter GmbH in Practical Metallography
- Vol. 62 (9-10) , 714-731
Deep learning-based precipitate quantification in STEM images of complex steel microstructures
Understanding precipitate evolution in microalloyed high-strength low-alloyed (HSLA) steels is essential for optimizing their processing and mechanical properties. In this work, we present an automated workflow for precipitate quantification in scanning transmission electron microscopy (STEM) images of niobium and titanium HSLA steels, combining deep learning-based semantic segmentation with conventional image processing techniques. Among the different STEM image contrasts tested, the high-angle annular dark-field-based model delivered the best segmentation performance, achieving a mean intersection over union of 0.8111 and a deviation of only 5.37 % in the measured mean particle diameter. The workflow’s robustness and generalizability were demonstrated on unseen images of different HSLA steels, with particle counts and size distributions aligning well with expectations based on the respective alloying concepts.
- Open Access
- 10 April 2024
- Published by Springer Nature in MRS Communications
- Vol. 14 (3) , 363-371
Enhancing machine learning classification of microstructures: A workflow study on joining image data and metadata in CNN
In view of the paradigm shift toward data-driven research in materials science and engineering, handling large amounts of data becomes increasingly important. The application of FAIR (findable, accessible, interoperable, reusable) data principles emphasizes the importance of metadata describing datasets. We propose a novel data processing and machine learning (ML) pipeline to extract metadata from micrograph image files, then combine image data and their metadata for microstructure classification with a deep learning approach compared to a classic ML approach. The ML model attained excellent performances with and without metadata and bears potential for performance improvement of further use cases within the community. Graphical abstract
- 10 April 2024
- Published by Springer Nature in MRS Communications
- Vol. 14 (3) , 363-371
Open AccessEnhancing machine learning classification of microstructures: A workflow study on joining image data and metadata in CNN
In view of the paradigm shift toward data-driven research in materials science and engineering, handling large amounts of data becomes increasingly important. The application of FAIR (findable, accessible, interoperable, reusable) data principles emphasizes the importance of metadata describing datasets. We propose a novel data processing and machine learning (ML) pipeline to extract metadata from micrograph image files, then combine image data and their metadata for microstructure classification with a deep learning approach compared to a classic ML approach. The ML model attained excellent performances with and without metadata and bears potential for performance improvement of further use cases within the community. Graphical abstract
- 1 January 2007
- Published by Taylor & Francis in SAR and QSAR in Environmental Research
- Vol. 18 (1-2) , 155-168
Discriminating toxicant classes by mode of action: 3. Substructure indicators
Decision support for selecting suitable QSARs for predictive purposes is suggested by a stepwise procedure: The first tier pre-filters the compounds based on substructure indicators for baseline versus excess toxicity. This step, if sufficiently conservative, discriminates chemicals, whose toxicity can be reliably estimated from their log K OW from those, that require further classification by biological and chemical domain. A test set of 115 chemicals from 9 different MOA classes was used to compare the discriminatory power of several classification schemes based on substructure indicators. Performance, evaluated by contingency table statistics, is varied and no single scheme provides sufficient applicability and reliability for pre-filtering chemical inventories. Major improvements are feasible with combined use of three classification schemes: assignments of baseline toxicants are protective, recognition of excess toxicants is acceptable and applicability range increases favourably.
- 1 January 2007
- Published by Taylor & Francis in SAR and QSAR in Environmental Research
- Vol. 18 (1-2) , 155-168
Discriminating toxicant classes by mode of action: 3. Substructure indicators
Decision support for selecting suitable QSARs for predictive purposes is suggested by a stepwise procedure: The first tier pre-filters the compounds based on substructure indicators for baseline versus excess toxicity. This step, if sufficiently conservative, discriminates chemicals, whose toxicity can be reliably estimated from their log K OW from those, that require further classification by biological and chemical domain. A test set of 115 chemicals from 9 different MOA classes was used to compare the discriminatory power of several classification schemes based on substructure indicators. Performance, evaluated by contingency table statistics, is varied and no single scheme provides sufficient applicability and reliability for pre-filtering chemical inventories. Major improvements are feasible with combined use of three classification schemes: assignments of baseline toxicants are protective, recognition of excess toxicants is acceptable and applicability range increases favourably.