Subject Training

The project aims to develop an intelligent welding assistance system based on multisensor signal analysis and machine learning methods, in order to automatically detect welding defects and improve process quality in real time. This tool is also intended to support the learning and refinement of good welding practices, both for apprentices and for the continuing training of experienced welders. The objective is to design a system capable of collecting and analyzing heterogeneous data in real time and providing the welder, during bead execution, with relevant information on the quality of the gesture and the process. The project therefore relies on the integration of physical sensors, imaging acquisition devices, visual, auditory, or haptic feedback systems, as well as advanced signal processing and data analysis algorithms to monitor and optimize the welding process. Arc welding generates complex, noisy, and highly non-stationary time-series data from multiple sensors (voltage, current, imaging, etc.). The internship will mainly focus on the software aspect of the project, and more specifically on the design and implementation of algorithms for processing and analyzing the signals and data acquired during welding. Particular attention will be given to data-driven approaches, including relevant feature extraction, supervised and unsupervised learning, as well as anomaly detection methods, with the aim of automatically identifying characteristic signatures of welding defects. This work follows an initial collaboration with the CRAN, which demonstrated the links between porosity and welding voltage. The work will take place within the company Alfa Laval, which provides the industrial expertise required for deploying the data acquisition system, and at the CRAN, which will ensure scientific supervision on data processing aspects.