Scope of Work
In mechanical tests such as tensile, creep, and bending creep (cantilever) experiments, the deformation behavior of specimens under load is investigated. For modern image based measurement techniques using speckle patterns, the accurate correspondence between undeformed and deformed surface points is crucial. The aim of this master’s thesis is to develop and evaluate neural networks that learn these mappings (displacement/registration) and subsequently enable automated data evaluation. Furthermore, benchmarking of this approach with respect to accuracy and runtime against classical methods (e.g., DIC) may be performed.
The following challenges shall be addressed:
Analysis of tensile and creep tests with speckle pattern based surface structures
- Preparation of available image/measurement data and definition of a ground truth setup
- Development of an ML based approach for displacement field prediction (image to image / flow)
- Evaluation of existing concepts/repositories as a starting point:
- USL_DIC (GitHub): https://github.com/cxn304/USL_DIC
- Reference article: https://doi.org/10.1016/j.optlastec.2024.111414
CNN based approaches: https://arxiv.org/ftp/arxiv/papers/2110/2110.13720.pdf
- Exploratory: use of self supervised image representations (e.g., DINO) for robust correspondences
- Optional: “fingerprint” concept for robust feature identification
- Possible evaluation:
- Baseline setup: classical DIC / analytical methods
- Metrics: displacement RMSE, correlation with respect to ground truth, speedup factor (runtime), robustness to noise/illumination
Candidate Profile
- Master candidate in Computer Engineering, Computer Science, Materials Science, Physical Engineering Sciences, or a comparable program
- Very good programming skills in a scripting or analysis language (Python, MATLAB); experience with typed programming languages (e.g., C++) advantageous
- Basic knowledge of software engineering (e.g., design patterns, testing, version control) desirable
- Knowledge in digital signal/image processing and machine learning advantageous
- Interest in supporting scientific projects; experimental skills and experience with measurement equipment are an advantage
- Ability to work independently, team skills, good German and English skills
We Offer
- Current and application oriented insights into exciting topics in research and development, primarily in the field of non destructive testing
- Intensive technical supervision and a collegial working atmosphere
- Opportunity to contribute your own ideas and develop creative, interdisciplinary solution strategies
BAM Divisions
eScience Unit (executing & supervising)
Coordinates and advises on digitalization and data driven science across all research areas of BAM. Core topics include communication and coordination in interdisciplinary digital projects, data science & AI/ML, optimization and data mining, internal research data management, reference data structures, support for software development and prototyping for scientific applications, and support of projects with a data science focus.
Metallic High-Temperature Materials (application domain)
Research on process–structure–property relationships of novel high temperature alloys. Mechanical testing under service relevant conditions (including tensile/creep tests, multiaxial fatigue, thermomechanical loading paths) from room temperature to high temperatures (up to 1200 °C), crack propagation at elevated temperature, online monitoring (electrical potential method, digital image correlation, thermography), and microstructural analyses to identify damage mechanisms. The material spectrum includes additively manufactured alloys, superalloys, and high entropy alloys.
Application
- Start: as soon as possible
- Duration: 6 months
- Location/Hybrid: BAM UE
- Application documents: CV, current transcript of records, certificate of enrollment, short motivation pitch (5–7 sentences), preferred start date
- Contact: Alexander Kister, VP.1, Email: alexander.kister@bam.de