|
Lecturer(s)
|
|
|
|
Course content
|
Week 1: Introduction to the materials data ecosystem - Overview of the materials data landscape: experimental vs. computational, structured vs. unstructured. Introduction to the major databases (Materials Project, AFLOW, OQMD, CSD). Week 2: Preparing data for materials research - Creating a dataset for hydrogen embrittlement, using data from the literature on steel composition and embrittlement indices (EI). Week 3: Predicting material properties - Predicting hydrogen embrittlement in steels. Using the data set from Week 2, students will train models to predict embrittlement index based on steel composition and process parameters (e.g., hydrogen pressure, yield stress). They will evaluate and compare the performance of the models. Week 4: Mapping High Entropy Alloys - High Entropy Alloys (HEA). Using an extensive dataset of HEA composition and properties from the AFLOW database, students will visualize the high-dimensional composition space in 2D/3D identifying distinct families of alloys based on their properties, which can reveal non-trivial groupings. Week 5: Predicting hydrogen embrittlement - Creating a simple neural network. Students will build and train a basic neural network for EI prediction to re-solve the hydrogen embrittlement prediction problem from Week 3 and compare its performance to classical models. Week 6: Classification of defects in castings and forgings - Microstructure analysis, phase identification and defect detection in castings, forgings and after heat treatment. Week 7: Defect Classification in 3D Printing - Microstructure analysis, phase identification and defect detection in additive manufacturing to classify images into "defect" and "defect-free". Ability to compare defect types, sizes and frequencies of additive and standard material manufacturing methods (see week 6). Week 8: In-situ quality control capabilities in 3D printing - Locating defects in the powder bed of additive manufacturing. Using an annotated dataset from additive manufacturing, students will train a model to draw bounding boxes around specific defects (e.g., porosity, cracks, delamination) in images of each layer of a 3D print. Week 9: HEA Alloys II - Constraints on manual flag creation for crystal structures. The concept of representing a crystal as a graph where atoms are nodes and bonds are edges. Students apply this technique to HEA structures from the AFLOW database. Week 10: Predicting Alloy Properties from Atomic Structure (HEA III) - Predicting the formation energy of HEAs. Students will train a model on a subset of the Materials Project or AFLOW database to predict the formation energy of complex alloys, reiterating a key result of the original paper. Week 11: Discovering new/suitable materials for hydrogen storage - Screening new materials for hydrogen storage. Students apply their practiced model from Week 10 to an extensive list of candidate materials from the Materials Project or similar database. They will predict stability and/or storage capacity for thousands of compounds and identify a list of the best performing, unexplored candidates. Week 12: Mining Knowledge from the Materials Science Literature - Automated data extraction from scientific articles. Students will be given a set of images (e.g., stress-strain curves, phase diagrams) from articles on hydrogen embrittlement. They write a detector to extract raw key data points from these images, creating a structured dataset from unstructured sources. Week 13: Summary of Semester 1.
|
|
Learning activities and teaching methods
|
- Contact hours
- 52 hours per semester
- Undergraduate study programme term essay (20-40)
- 38 hours per semester
- Preparation for an examination (30-60)
- 30 hours per semester
|
| prerequisite |
|---|
| Knowledge |
|---|
| Basic knowledge of crystallography, thermodynamics, and mechanical properties of materials (necessary for interpreting results). Knowledge of linear algebra and statistics. User knowledge of Python and data processing libraries (NumPy, Pandas) is required for working with computational tools. |
| Skills |
|---|
| Analytical thinking and the ability to connect the physical essence of a problem with a data model. |
| Competences |
|---|
| Ability to learn independently and solve problems based on a framework assignment. |
| learning outcomes |
|---|
| Knowledge |
|---|
| Explains the potential of data-driven approaches ("fourth paradigm") for accelerating the development of new materials and the specifics of working with small and sparse material data. Understands the selection of appropriate machine learning algorithms for different types of material tasks (e.g., regression for predicting yield strength, classification for phase identification). Understands the principles of advanced architectures for material structure analysis: convolutional neural networks (CNN) for microstructure image analysis and graph neural networks (GNN) for crystal lattice representation. Is familiar with methods for virtual screening of new compounds (e.g., for energy storage) and the possibilities of automated data mining from scientific literature. Knows the criteria for evaluating the reliability of predictive models in the context of material safety and quality. |
| Skills |
|---|
| Create material datasets by extracting and cleaning data from public databases (Materials Project, AFLOW) or literature, prepared for training property models. Design and train predictive models for solving engineering problems, specifically for predicting mechanical properties (e.g., hydrogen embrittlement of steels) based on composition and processing. Apply computer vision methods for automatic detection and classification of manufacturing defects (pores, cracks) in additive manufacturing and conventional technologies. Use graph neural networks to predict the stability and energy of complex alloys (e.g., HEA) directly from their atomic structure. Visualize and interpret multidimensional data on the composition and properties of materials to discover new relationships and families of materials. |
| Competences |
|---|
| N/A |
| Apply existing AI/ML models to new but well-defined material problems. Critically assess the suitability of different predictive models for a given type of material data (numerical data, images, graphs). Clearly present the results of their data analysis and modelling. |
| teaching methods |
|---|
| Knowledge |
|---|
| Interactive exercises linking AI theory with examples from materials practice. Demonstration of real-world applications (e.g., examples of microstructure analysis). |
| Skills |
|---|
| Hands-on exercises on real data sets (steel, HEA alloys, 3D printing). Project-based teaching simulating a research task (from raw data to predicting a new material). |
| Competences |
|---|
| Independent study of recommended materials. Team consultations (optional for projects). |
| assessment methods |
|---|
| Knowledge |
|---|
| Ongoing tests to verify understanding of the principles of materials informatics. Final exam. |
| Skills |
|---|
| Evaluation of practical tasks (e.g., success rate of steel property prediction, defect detection accuracy). Defense of a semester project focused on a specific material problem. |
| Competences |
|---|
| Ability to solve problems, Present the results of smaller tasks. |
|
Recommended literature
|
-
Haiyang Yu, Andrés Díaz, Xu Lu, Binhan SunYu Ding, Motomichi Koyama, Jianying He, Xiao Zhou, Abdelali Oudriss, Xavier Feaugas, Zhiliang Zhang. Hydrogen Embrittlement as a Conspicuous Material Challenge?Comprehensive Review and Future Directions. 2024.
-
Kamal Choudhary & Brian DeCost. Atomistic Line Graph Neural Network for improved materials property predictions. 2021.
-
Kefan Chen, Peilei Zhang, Hua Yan, Guanglong Chen, Tianzhu Sun, Qinghua Lu, Yu Chen & Haichuan Shi. A review of machine learning in additive manufacturing: design and process. 2024.
-
Keith T. Butler, Felipe Oviedo, Pieremanuele Canepa. Machine Learning in Materials Science. 2022.
-
Mohamed Yasin Abdul Salam, Enoch Nifise Ogunmuyiwa, Victor Kitso Manisa, Abid Yahya, Irfan Anjum Badruddin. Effect of fabrication techniques of high entropy alloys: A review with integration of machine learning. 2025.
|