26 Novembre – Thesis defense - Valentine Cazaubon
10 h30 Amphi 300 - Bâtiment ESTIA (Bidart)
Additive manufacturing process by molten wire deposition: model, methods and strategies for the correction of mechanical part defect.
Metal additive manufacturing is a group of manufacturing processes, including the process of deposition by laser melting of wire, LMD-w in English. With these processes, it is possible to manufacture, correct or add new functions to mechanical parts. Within the framework of our research work, we have chosen to focus on the correction of mechanical part defects by metallic additive manufacturing. The objective of this research is to help in the choice of the machine parameters adapted to a problem of correction of part by adding material. For this purpose, we have developed a behavioral model integrating the complexity of the implementation of the LMD-w process related to multiphysical and multiscale phenomena. For the construction of this model, we have conducted three experimental campaigns consisting in manufacturing beads - elementary geometry of additive manufacturing - and then analyzing them with an optical instrument previously validated. This model allowed us to establish relationships between the input parameters studied - laser power, wire feed speed, tool displacement speed - and the geometric outputs - bead and dilution height, bead and dilution width, contact angle and substrate displacement -. Therefore, a knowledge model was developed. It consists of a neural network and a numerical simulation model of the LMD-w process. These two sub-models communicate with each other thanks to the generation of a mesh of the geometry to be simulated, at the output of the neural network. Finally, we have developed an optimization algorithm based on the principle of dominance in the Pareto sense to complete our proposal to help in the choice of machine parameters adapted to a correction problem. At the output of this algorithm, the data are visualized on an interactive graph called "parallel coordinates plot". Our work allows us to simulate optimal solutions of four correction strategies for mechanical parts with the LMD-w process, to explore the design space constituted by these optimal solutions and to select a "preferred" solution with the help of an interactive visualization of the performance and decision spaces.