Machine learning

DL Segmentation of Lath Bainite – LOM*

DL Segmentation of Lath Bainite – SEM*
Benefits of Machine Learning (ML)
- Offers promising new potentials in microstructure analysis
- Analysis of big and high-dimensional tabular data & sophisticated image analysis
- Automation, objectivity, reproducibility
- Automated and more efficient processing of labor-intensive standard tasks
- Basis for improved analysis of complex microstructures, which previously was not possible
How we approach ML
- Do not just throw images into ML algorithms
- View ML in a holistic approach: also consider all the steps that lead to acquiring micrographs
- Pay special attention to the ground truth assignment for supervised ML
- Controlling and understanding all material science related aspects that provide greater leverage for success than optimization of ML parameters
Where our ML approaches were successfully applied
- Deep Learning (DL) segmentation of lath bainite in multi-phase steels
- Object-wise ML classification of bainitic subclasses in dual-phase-steels
- Segmentation of prior austenite grains in light optical micrographs
- Classification of fracture surfaces
- Classification of non-metallic steel inclusions based on SEM images with BSE contrast
What we offer
- Feasibility studies of ML segmentation and classification, including generation of data and annotations
- Approaches for objective and reproducible ground truth assignment, e.g., correlative microscopy
- Holistic approach: steps and parameters that must be considered to ensure a sustained successful implementation of ML-based analysis
- Exploratory data analysis by unsupervised ML

DL classification of inclusions in SEM images

Exploratory data analysis (unsupervised)

DL reconstruction of prior austenite grains
Contact for questions

Dr.-Ing Dominik Britz
Deputy Head MECS Saarbrücken

Adrian Thome, M. Sc
Project Leader
Further areas of application
Correlative Microscopy
A correlative characterization of microstructures serves as a benchmark for automated microstructure recognition and as a starting point for machine learning approaches.
Triboelectrical characterization
Complementary to our microstructure analysis, we are able to perform a wide variety of triboelectric measurements (e.g. electrical resistivity and coefficient of friction during mating tests) under controlled conditions.
Failure analysis
In the area of failure analysis, in addition to identifying and analyzing causes of fracture, we are also able to fully characterize components on both the macro and micro scales.
Antimicrobial surfaces
Our antimicrobial metallic surfaces have already been used in 3 research projects with ESA & NASA on the ISS.
You are interested in working with us?
Feel free to contact us! We look forward to talking to you and finding out how we can help you with your project.