WG4

Explainable Machine Learning – MecaNano Tutorial Series

This tutorial focuses on explainable machine learning for materials mechanics and nanomechanical testing. The session discusses why explainability is essential when machine-learning models are used to analyse experimental datasets, where the results must remain physically meaningful and scientifically defensible. It introduces strategies to understand model decisions, identify relevant features, evaluate model reliability, and avoid black-box conclusions that cannot be connected to the underlying material behaviour or experimental conditions.

Tutorial Slides

The following documents can be read directly from this page but are protected from downloading.

1. Overview of Gaussian Processes

500012

Machine learning bases and advanced applications for nanoindentation data analysis – MecaNano Tutorial Series

This tutorial introduces the foundations of machine learning for nanoindentation and nanomechanical data analysis. The session covers the basic concepts of supervised and unsupervised learning, feature extraction from indentation curves, clustering and classification of indentation datasets, analysis of high-throughput nanoindentation maps, and advanced workflows based on the full load-displacement curve. It discusses how data-driven methods can support phase identification, detection of anomalous curves, interpretation of mechanical populations, and integration with correlative microstructural information.

Tutorial Slides

The following documents can be read directly from this page but are protected from downloading.

Slides Deck

EXPERIMENTAL NANOMECHANICS at the EMMC19 conference

Prof. Christoph Kirchlechner and Jon Molina are organizing a symposium on “EXPERIMENTAL NANOMECHANICS” at the EMMC19 conference (19thEuropean Mechanics of Materials Conference), which will take place on May 29th-31st, 2024 in Madrid, Spain (http://emmc19.org/symposium-s13.html). This symposium aims to bring together the research community in micro- and nano-mechanics, particularly those adopting interdisciplinary approaches, with the objective of gaining insights into the … Read more

CNNs for damage assessment in steels – MecaNano Tutorial

This tutorial is part of the program of the 2nd MecaNano Workshop on Machine Learning for Micro- and Nano-Mechanics.

Understanding the behaviour of steel under applied stress is fundamental for predicting its performance in real applications. In this tutorial, we explore how to characterise different types of damage in dual-phase steel using panoramic electron microscopy images combined with advanced analysis techniques. The data and scripts used here are openly available, making it possible to reproduce and extend the work.

Repository with data and code: https://git.rwth-aachen.de/teaching_uk/matphysdesign/damageclassification

1. Introduction

When tensile stress is applied to steel, multiple microstructural damage mechanisms can be activated. Identifying and distinguishing these mechanisms is key for improving material design and predicting lifetime under load.

In this tutorial, we focus on the following damage types:

  • Martensite Cracking – brittle cracks forming within martensite islands.

  • Notch Effect – stress concentration at geometric irregularities.

  • Interface Decohesion – debonding at ferrite–martensite interfaces.

Additionally, we account for two major challenges:

  • Manufacturing-related artefacts – foreign particles embedded in the sample that mimic damage sites.

  • Imaging artefacts – shadows and contrast distortions caused by the 3D geometry of the specimen under the electron beam.

Our starting point is a panoramic electron microscopy image of a dual-phase steel specimen, from which damage sites will be identified and classified.

2. Workflow Overview

The repository provides both data and code to reproduce the analysis. The general steps are:

  1. Acquire panoramic image of the steel specimen using a scanning electron microscope (SEM).

  2. Pre-process the image to remove artefacts (beam shadows, noise).

  3. Detect candidate damage sites (cracks, voids, notches).

  4. Classify damage mechanisms into martensite cracking, notch effect, or interface decohesion.

  5. Exclude false positives from inclusions or artefacts.

  6. Visualise and validate results on the large-area image.

The scripts in the repository automate parts of this workflow, and can be adapted for new datasets.

3. Related Work

This tutorial builds upon and connects to recent advances in large-area imaging and machine learning for microstructural damage analysis:

  • Large-area SEM image analysis with deep learning
    Kusche C, Reclik T, Freund M, Al-Samman T, Kerzel U, Korte-Kerzel S (2019)
    Large-area, high-resolution characterisation and classification of damage mechanisms in dual-phase steel using deep learning.
    PLoS ONE 14(5): e0216493.

  • 3D characterisation of damage in dual-phase steel
    Setareh Medghalchi, Ehsan Karimi, Sang-Hyeok Lee, Benjamin Berkels, Ulrich Kerzel, Sandra Korte-Kerzel (2023)
    Three-dimensional characterisation of deformation-induced damage in dual phase steel using deep learning.
    Materials & Design, Volume 232, 112108.
    https://doi.org/10.1016/j.matdes.2023.112108

4. Data and Code

All original data and scripts for reproducing this tutorial are openly available on GitLab: https://git.rwth-aachen.de/teaching_uk/matphysdesign/damageclassification

Gaussian Process Regression — MecaNano Tutorial Series

This tutorial introduces Gaussian Process Regression (GPR), a non-parametric probabilistic approach widely adopted for interpolation, regression, and uncertainty quantification in materials science. The session explores both 1D and 2D inference, implemented with pyro.contrib.gp and PyTorch, with examples crafted for clarity and practical use.

Tutorial Slides

The following documents can be read directly from this page but are protected from downloading.

1. Overview of Gaussian Processes

2. Univariate GP Regression

3. Bivariate GP Regression

Note: Download buttons have been disabled for the embedded documents to protect the tutorial content. Please contact the authors for access.

MecaNano Summer School 2025

The long awaited MecaNano Summer School 2025 is now fully organized to present new topics related to Mechanics at the Nanoscale! The summer school will be held 28. July to 1. August in Kassel, Germany home to the UNESCO mountain park and elected happiest city of Germany in 2024. The organizers (B. Merle, C. Kirchlechner, … Read more

2nd MecaNano Workshop on Machine Learning for Micro- and Nano-Mechanics

The 2nd MecaNano Workshop on Machine Learning for Micro- and Nano-Mechanics will take place on September 4–5, 2025, at ELTE Eötvös Loránd University, Faculty of Science, Budapest. Following the success of the first edition, this workshop continues to explore the integration of machine learning techniques in small-scale mechanical testing and materials science, with a focus … Read more

New book by MecaNano member on Machine Learning

MecaNano Member and WG4 co-lead, Stefan Sandfeld, has published a new book titled “Materials Data Science — Introduction to Data Mining, Machine Learning, and Data-Driven Predictions for Materials Science and Engineering.” The book is aimed for Material Science students in their first Masters of Science semesters and PhD students or scientists who haven’t had much … Read more

First topical workshop on Machine Learning (WG4) – Feb. 22-23, 2024 – ABSTRACT SUBMISSION OPEN

The abstract submission is still open for our 2-day event related to the WG4 “Machine Learning” here: https://mecanano-wg4-24.sciencesconf.org/ The workshop will be held on 22-23 February 2024 in Louvain-la-Neuve, Belgium. (Note: this is a different city to Louvain!) The community is rather diverse and has a large range of experienced in the use of ML: … Read more