Subject Training

Alfa Laval Golbey is a key player in cryogenics, specialized in the design and manufacturing of brazed heat exchangers used in the gas industry (air separation, LNG, hydrogen, petrochemicals, etc.). A previous internship led to the development of an initial solution for counting compactness defects in brazed joints, based on color image processing. While this approach demonstrated a proof of concept, it remains sensitive to industrial variations (lighting, surface defects, acquisition noise), which limits its use in fully automatic mode. At the same time, new acquisition systems, such as polarimetric imaging or eddy current inspection, could eliminate the need for dye penetrant testing and improve inspection quality.

Internship Objectives The internship aims to strengthen the robustness and automation of brazed-joint analysis. 1. Evaluation and improvement of the existing tool -Test current parameters on a wide range of industrial images. -Optimize processing methods (filtering, thresholding, segmentation). Introduce, if needed, machine-learning approaches to enhance detection (classification, segmentation with lightweight deep-learning models) when classical processing reaches its limits. 2. Integration of a suitable acquisition system -Investigate the contribution of other modalities (polarization, eddy currents, etc.) to improve defect detection. -Compare their performance for internal-defect detection. -Develop the associated algorithm, including automatic defect identification and counting.

Researched profile: Master's student (M2) specializing in signal/image processing, data analysis, machine learning, applied mathematics, or embedded systems. Strong programming skills (Python or MATLAB). Knowledge of signal and image processing algorithms. Interest in industrial applications of new technologies.

Contacts For applications or additional information, please contact: Alfa Laval Golbey R&D: gaetan.bergin@fivesgroup.com ; thierry.mazet@fivesgroup.com CRAN Nancy: sebastian.miron@univ-lorraine.fr ; julien.flamant@univ-lorraine.fr ; david.brie@univ-lorraine.fr