Master Thesis: AI for Real-Time Microstructure Monitoring Using Laser Ultra

Swerim AB; Stockholms kommun; On-site; SEK 45,000 - 55,000; Full time

Job summary

Swerim AB in Stockholm invites you to join the NAICE project as a master thesis student in Laser Ultrasonics data science. You will apply machine learning to LUS data from hot strip mill conditions, focusing on grain size prediction and microstructure classification.

The work involves pre-processing high-resolution A-scan data, surveying AI architectures, developing models, and benchmarking performance for real-time applicability.

Qualifications

• A background in computer engineering, engineering physics, materials science, applied mathematics, or similar.
• Strong interest in machine learning and signal processing.
• Experience with Python/Matlab and ML frameworks.
• Basic understanding of wave physics, ultrasonics, or materials science.
• Analytical mindset and ability to work with complex datasets.

Responsibilities

• Survey AI/ML architectures for ultrasonic time-series data.
• Pre-process and structure high-resolution LUS datasets for model training.
• Develop and implement selected AI models.
• Perform benchmark comparisons on accuracy, robustness, and efficiency.
• Analyze model interpretability in relation to physical phenomena.

Skills

Python

Matlab

ML frameworks

Signal processing

Analytical mindset

Education

Master's student in related field

Tools

Python

MATLAB

Job description

Project Description

Laser Ultrasonics (LUS) is a powerful non‑contact technique for in‑situ characterization of metallic materials at elevated temperatures. In hot strip mill environments, LUS enables real‑time monitoring of grain size evolution through analysis of ultrasonic wave attenuation and velocity, providing critical insight into recrystallization, grain growth, and phase transformations. This information is highly valuable for advanced process control strategies aimed at improving product quality, consistency, and resource efficiency.

The NAICE project (Non‑destructive AI enhanced online grain size evaluation) addresses a key challenge in fossil free steel production: today’s steel mills lack real time information about the microstructure that governs material quality, leading to unnecessary energy use, scrap, and slow process adjustments. By upgrading the world’s only industrial laser ultrasonic grain size gauge and collecting a unique large dataset, the project will train models to enable AI enhanced measurement of the grain size and microstructure in real‑time. This enables smarter process control, higher yield, and reduced emissions.

This thesis is the first of two planned master thesis works. The first will evaluate the best options using historical data sets, whereas the second will focus on using the full dataset and optimizing the model for industrial implementation.

If you want to know more about the LUS technology and the installation of the world’s only grain size gauge there is a recorded webinar here: https://www.youtube.com/watch?v=ZOpXUAPhEv0&t=665s

Thesis Scope

The thesis will focus on the application of machine learning and AI techniques to Laser Ultrasonic (LUS) data acquired in a hot strip mill environment. The available dataset consists of:

• Raw LUS A‑scan signals with multiple backwall echoes (20 µs duration, 1 ns resolution)
• Grain size values calculated using an established conventional algorithm (serving as reference values)
• The primary objective is to develop and evaluate AI‑based approaches for:
  • Grain size prediction directly from raw laser ultrasonic signals
  • Microstructure state classification (e.g., recrystallized fraction, pancaking/aspect ratio)

A key part of the work will be to survey suitable AI architectures, such as time‑series models, deep learning approaches, and alternative signal‑processing‑informed models, and assess their applicability to the LUS data.

Tasks

• Survey and critically assess relevant AI/ML architectures for ultrasonic time‑series data
• Pre‑process and structure high‑resolution LUS A‑scan datasets for model training
• Develop and implement selected AI models
• Perform benchmark comparisons between different approaches with respect to:
• Prediction accuracy
• Robustness to signal variability
• Computational efficiency and suitability for real‑time deployment
• Analyze model interpretability in relation to known physical phenomena

Expected Outcomes

• A structured comparison of candidate AI approaches for LUS‑based grain size prediction
• Prototype models demonstrating feasibility on the historical dataset
• Recommendations for model selection in future industrial implementation

Desired Qualifications

• A background in computer engineering, engineering physics, materials science, applied mathematics, or similar
• Strong interest in machine learning and signal processing
• Experience with Python/Matlab and ML frameworks
• Basic understanding of wave physics, ultrasonics, or materials science (meriting but not required)
• Analytical mindset and ability to work with complex datasets

Additional Information

• The work will be carried out in close collaboration with experts in Laser Ultrasonics and industrial process monitoring, providing a unique opportunity to contribute to cutting‑edge development of AI‑driven process control in steel manufacturing.

Project time

The project is intended for a master thesis (30hp), starting in spring 2027 or can be mutually decided through negotiations.

Further information

This project is intended to be performed at Swerim in Stockholm. Swerim rewards the student with 50 000 SEK for an approved master thesis (30hp).

Contacts

• Mikael Malmström, [e-mail]
• Hampus Wikmark Kreuger, [e-mail]