Machine Learning in Materials Processing & Characterization
Application-focused course on ML for experimental materials data, from images and spectra to process signals.
1 Machine Learning in Materials Processing & Characterization
5th Semester – 5 ECTS · 2h lecture + 2h exercises per week
Coordinated with “Mathematical Foundations of AI & ML” (MFML)
and “Materials Genomics” (MG)
1.1 Synergy Map
- MFML provides the mathematical spine: loss functions, neural networks, generalization, uncertainty, Gaussian Processes.
- This course (ML-PC) applies these concepts to experimental data: images, spectra, and processing signals.
- Materials Genomics focuses on crystal structures, databases, and discovery.
ML-PC is therefore application-driven, not algorithm-driven.
1.2 Companion books
- Sandfeld (2024): Materials Data Science
1.3 Week-by-Week Curriculum (14 weeks)
1.3.1 Unit I — Experimental Data as a Learning Problem (Weeks 1–3)
1.3.1.1 Week 1 – What makes materials data special?
Lecture: Tuesday, 14.04.2026, 14:15-15:45 | Exercise: Thursday, 16.04.2026, 16:15-17:45
Slides: Open
- Types of experimental data: micrographs, EBSD, EDS, EELS, XRD, process logs, thermal histories.
- PSPP (Processing–Structure–Property–Performance) as a data dependency graph.
- Why ML failure modes are common in experimental science.
Summary:
- Transition from physics-based to data-driven modeling
- Experimental data challenges: multi-modal, high acquisition cost, sparse
- PSPP (Processing → Structure → Property → Performance) as a data dependency graph
- Data scales and measurement uncertainty
- CRISP-DM workflow adapted for scientific labs
Exercise:
Inspect real microscopy and process datasets; identify sources of bias and noise.
1.3.1.2 Week 2 – Physics of data formation
Lecture: Tuesday, 21.04.2026, 14:15-15:45 | Exercise: Thursday, 23.04.2026, 16:15-17:45
Slides: Open
- Image and signal formation in characterization: resolution, contrast, artifacts.
- Sampling, aliasing, noise as physical priors (not preprocessing tricks).
- Relation to MFML refresher on PCA and covariance.
Summary:
- Physical signal formation as a learning prior
- Resolution, noise, sampling as physical (not algorithmic) constraints
- PCA and SVD for low-dimensional structure in high-dimensional data
Exercise:
Fourier inspection of micrographs; effects of sampling and filtering.
1.3.1.3 Week 3 – Data quality, labels, and leakage
Lecture: Tuesday, 28.04.2026, 14:15-15:45 | Exercise: Thursday, 30.04.2026, 16:15-17:45
Slides: Open
- Annotation uncertainty and inter-annotator variance.
- Train/test leakage in materials workflows.
- Why “good accuracy” often means a broken pipeline.
Summary:
- Measurement chain → data cleaning: missing values, outliers, duplicates (“fix at source”)
- Transformation toolbox: centering, min–max / z-score scaling, non-dimensionalization, log, differentiation, FFT, triggering
- Labels and uncertainty: inter-annotator variance, probabilistic labels, Bayesian view (priors, likelihoods, posteriors)
- Bias–variance tradeoff with parsimony and regularization
- Data leakage in materials workflows: pre-processing, temporal, group/spatial
- Validation: holdout, K-fold, LOOCV, stratified
- Error measures:
- Regression: MAE, MSE, RMSE, \(R^2\)
- Classification / segmentation: confusion matrix, precision/recall, F1/Dice, IoU, categorical cross-entropy
Exercise:
Construct a deliberately flawed ML pipeline and diagnose its failure.
1.3.2 Unit II — Representation Learning for Microstructures (Weeks 4–6)
(Aligned with early neural networks in MFML)
1.3.2.1 Week 4 – From classical microstructure metrics to learned representations
Lecture: Tuesday, 05.05.2026, 14:15-15:45 | Exercise: Thursday, 07.05.2026, 16:15-17:45
Slides: Open
- Grain size, phase fractions, orientation maps.
- Limits of hand-crafted microstructure features.
- Transition to learned representations.
Summary:
- Classical stereological metrics (grain size, phase fractions) and their limits
- Transition to learned representations
- The artificial neuron: weights, biases, non-linear activations
- Multi-Layer Perceptrons (MLPs) as automatic feature learners
Exercise:
Compare classical features vs simple NN-based features for microstructure tasks.
1.3.2.2 Week 5 – Neural networks for microstructure images
Lecture: Tuesday, 12.05.2026, 14:15-15:45 | Exercise: Thursday, 14.05.2026, 16:15-17:45 (cancelled - Himmelfahrt)
Slides: Open
- CNN intuition: filters as structure detectors.
- Example tasks: phase segmentation, defect detection, porosity identification.
- Overfitting risks with small datasets.
Summary:
- Convolutional Neural Networks (CNNs) for materials characterization
- Hierarchical structure detectors: edges → textures → phase morphologies
- Filters and pooling; parameter efficiency vs. MLPs
- Case studies: phase segmentation, defect detection
- Practical challenges: high-resolution, noisy micrographs
Exercise:
Train a small CNN on microstructure images; analyze failure cases.
1.3.2.3 Week 6 – Data scarcity & transfer learning
Lecture: Tuesday, 19.05.2026, 14:15-15:45 | Exercise: Thursday, 21.05.2026, 16:15-17:45
Slides: Open
- Why materials datasets are small.
- Transfer learning from natural images vs self-supervised pretraining.
- When transfer learning helps—and when it does not.
Summary:
- Data scarcity as the materials informatics bottleneck
- Transfer learning from natural-image pretrained models
- Self-supervised pretraining as an alternative
- Data augmentation tailored to scientific data
- When cross-domain transfer succeeds vs. fails
Exercise:
Fine-tune a pretrained model; compare against training from scratch.
1.3.2.4 Week 7 – No lecture (26.05.2026, public holiday)
Cancelled — no ML-PC lecture or exercise takes place in Week 7 (Pfingstmontag / Pfingstdienstag public holidays, 25–26.05.2026). The time-series & process monitoring unit moves to Week 8 (02.06.2026); every later unit shifts one week, taking up the former Week-14 buffer slot.
1.3.3 Unit III — Learning from Processing Data (Weeks 8–9)
1.3.3.1 Week 8 – Time-series and process monitoring
Lecture: Tuesday, 02.06.2026, 14:15-15:45 (in class) | Exercise: Thursday, 04.06.2026, 16:15-17:45 (self-study — Fronleichnam public holiday)
Slides: Open
Self-study exercise: the Thursday slot is cancelled (Fronleichnam). The exercise is provided for independent work; a solution is released afterwards. The Tuesday lecture takes place in class.
- Processing signals: temperature cycles, AM melt pool signals, SPS, rolling.
- Regression and sequence models as surrogates.
- Relation to MFML concepts of generalization.
Summary:
- Time-series ML for process monitoring and prediction
- RNNs and LSTMs for sequential dependencies
- Preprocessing: signal smoothing, triggering on noisy logs
- Case studies: additive manufacturing, process stability
- Real-time anomaly detection from processing history
Exercise: Predict a process outcome from time-series data using regression or simple RNNs.
1.3.3.2 Week 9 – Inverse problems and process maps
Lecture: Tuesday, 09.06.2026, 14:15-15:45 | Exercise: Thursday, 11.06.2026, 16:15-17:45
Slides: Open
- Process → structure inverse problems.
- ML-guided process maps (e.g. AM laser power vs scan speed).
- Physics-informed vs unconstrained regression.
Summary:
- Inverse problems: target microstructure / performance → processing parameters
- Forward (causal) vs. inverse (often ill-posed, multi-valued)
- Physics-informed learning: physical transformations and constraints
- Process maps and process corridors for safe operating regions
Exercise: Construct a simple ML-based process map; compare constrained vs unconstrained models.
1.3.4 Unit IV — Characterization, Transformers, and Uncertainty (Weeks 10–12)
1.3.4.1 Week 10 – ML for characterization signals
Lecture: Tuesday, 16.06.2026, 14:15-15:45 | Exercise: Thursday, 18.06.2026, 16:15-17:45
Slides: Open
- Spectral data: XRD, EELS, EDS.
- Denoising, peak finding, dimensionality reduction.
- Using ML without destroying physical meaning.
Summary:
- Unsupervised ML on high-dimensional spectra (XRD, EDS, EELS)
- K-Means and t-SNE for phase identification and visualization
- Autoencoders: compressing spectra into a low-dimensional latent space
- Denoising and feature extraction at high throughput without losing physics
Exercise: Apply PCA/NMF to spectral datasets; interpret components physically.
1.3.4.2 Week 11 – Transformers for materials characterization
Lecture: Tuesday, 23.06.2026, 14:15-15:45 | Exercise: Thursday, 25.06.2026, 16:15-17:45
Slides: Open
- Why attention: long-range correlations beyond CNN receptive fields.
- Scaled dot-product attention and the Vision Transformer (ViT).
- Flash Attention for tractable long sequences.
Summary:
- Self-attention and the Vision Transformer (ViT) for materials imaging
- Flash Attention: long sequences without the L×L memory blow-up
- Applications: ViT on 4D-STEM diffraction; cross-attention across LPBF layer stacks
- Scaling alternatives (Mamba / state-space models) — and when not to reach for a transformer
Exercise: Apply a small ViT / attention model to a characterization dataset (e.g. 4D-STEM patches); compare against a CNN baseline.
1.3.4.3 Week 12 – Uncertainty-aware regression & Gaussian Processes
Lecture: Tuesday, 30.06.2026, 14:15-15:45 | Exercise: Thursday, 02.07.2026, 16:15-17:45
Slides: Open
- Aleatoric vs epistemic uncertainty in experiments.
- Gaussian Processes as uncertainty-aware surrogates.
- Exploration vs exploitation in experimental design.
- Connection to materials acceleration platforms.
Exercise: Compare GP regression and NN ensembles for a process-parameter problem.
1.3.5 Unit V — Physics, Trust, and Synthesis (Weeks 13–14)
1.3.5.1 Week 13 – Physics-informed and constrained ML — self-study (lecture cancelled)
Lecture cancelled (Tuesday, 07.07.2026) — delivered as required self-study. Exercise: Thursday, 09.07.2026, 16:15-17:45 takes place as scheduled.
The Tuesday lecture does not take place. This unit — physics-informed and constrained ML — is required self-study from the slide deck below. It remains examinable and is the unit that delivers the “integrate physical constraints into ML workflows” learning outcome. The Thursday exercise runs as normal.
Slides (required self-study): Open
- Embedding physical constraints into ML models.
- Penalty terms, soft constraints, hybrid approaches.
- Failure modes of unconstrained models.
Exercise: Train a constrained model for a processing or characterization task.
1.3.5.2 Week 14 – Integration, limits, and reflection
Lecture: Tuesday, 14.07.2026, 14:15-15:45 | Exercise: Thursday, 16.07.2026, 16:15-17:45
Slides: Open
- Explainability for experimental ML (CAMs, SHAP).
- Why ML fails in real labs.
- Where ML genuinely changes materials processing.
Exercise: Mini-project presentations and critical discussion.
1.4 Learning Outcomes
Students completing this course will be able to:
- Interpret materials processing and characterization data as learning problems.
- Build ML pipelines for microstructure analysis, process prediction, and spectral data.
- Understand the physics of data formation to avoid common ML pitfalls.
- Evaluate generalization, robustness, and uncertainty in experimental ML models.
- Apply Gaussian Processes and neural networks as surrogate models.
- Integrate physical constraints into ML workflows.
- Critically assess claims about ML in materials processing and characterization.
1.5 Lab Possibilities
- Microscopy datasets: noise, metadata, units, and artifacts.
- Fourier inspection of SEM/TEM images.
- Broken vs correct ML pipelines (data leakage case studies).
- Feature extraction vs learned representations.
- Fine-tuning pretrained CNNs on microstructures.
- Process–property regression with uncertainty.
- GP-based process maps.
- Spectral decomposition (NMF) of EELS/XRD data.
- ML-assisted autofocus or EBSD pattern classification.
- Multi-modal fusion of images, spectra, and process parameters.