From classical to modern: deep learning for metals and alloys

At MSE 2026 we walked the room through four stages of microstructure analysis, from grey-value thresholds to deep learning and prompt-based AI, using real images of brass, titanium, steel welds and powders. Our partner ADDITIVE gave a second lecture, and between sessions we ran live demos at the booth we share with them.

LecturePosterBooth

Book a free demo MIPAR →

What we showed

On stage

The lecture followed one argument in four steps.

Classical image analysisThe established pipeline of acquire, pre-process, threshold, measure and report. It works on high-contrast, well-prepared samples and struggles with twins, overlapping phases, noise and operator bias.
Deep learning in actionTwo case studies: twinned grains in brass and beta grains in titanium, both measured automatically where classical thresholding fails.
AI and SpotlightPrompt-based segmentation. Click a feature and Spotlight finds every similar one. A small network trained on 2–3 annotated images can drive the prompts on its own.
The modern approachClassical tools, deep learning and Spotlight combined in one no-code recipe, with batch processing and reports.

Twinned grains in brassDeep learning finds the real grain boundaries and ignores the twins. The grain map is ready for an ASTM E112 measurement.

Our MSE 2026 poster: MIPAR Spotlight for materials applications.

Takeaways

What changes when AI enters the workflow

SpeedFrom hours of manual work to seconds per image. Thousands of images can run overnight in a batch.

AccuracyDeep learning detects features classical algorithms miss, especially in low-contrast or noisy images.

ReproducibilityThe same result every time, whatever the operator, the instrument or the day of the week.

GeneralisationTrain on a few images. The model handles new samples without being reprogrammed for every batch.

Read the abstract as submitted

AI-Enabled Image Analysis for Quantitative Microstructural Characterization

Recent advances in machine learning are redefining how material microstructures are decoded from imaging data. As modern microscopy techniques generate increasingly complex and high-volume datasets, traditional manual and semi-automated analysis methods struggle to deliver the throughput, consistency, and statistical robustness required for reliable microstructural interpretation. AI-driven image analysis is emerging as a critical enabler for extracting quantitative insight from microstructures, allowing researchers to more effectively link grains, phases, and defects to material performance.

This presentation examines how deep-learning-enabled image analysis workflows can accelerate and standardize microstructural characterization across a range of materials systems. Using MIPAR as an example, we demonstrate how no-code, recipe-based pipelines support common yet challenging tasks such as grain size and shape analysis with twin boundary exclusion, phase segmentation in multiphase materials, crack and pore detection, and quantitative assessment of defect morphology and distribution. The integrated Spotlight deep learning engine enables efficient generation of training datasets directly from microstructural images, facilitating rapid development of accurate segmentation models without the need for custom algorithms or manual parameter tuning.

Through representative case studies, we highlight how precise segmentation and feature extraction improve statistical confidence and enable higher-throughput microstructural analysis. These examples illustrate how accessible AI-powered tools are helping to decode complex microstructures more reliably, supporting process optimization, failure analysis, and the broader transition toward data-driven materials characterization and design.

About the congress

MSE 2026, Materials Science and Engineering Congress

MSE is Germany's leading international congress for materials science and engineering. It takes place every two years in Darmstadt and covers biomaterials, characterisation, digital transformation, functional and circular materials, modelling and simulation, processing and structural materials.

Visit mse-congress.de →