| Περίληψη: | The manufacturing industry is heading from mass production to the mass customization era, depending on the dynamic customer needs. A major part of this transition is reconfiguring the production to adapt to this dynamic market. Laser welding is an important manufacturing process for industry, as not only is more accurate than any other welding process, it is also very fast. Simultaneously, the physics that govern the process, are numerous and complex, which makes the selection of correct process parameters difficult, when it alternates constantly. Simulating the process could be a valuable tool for the industry for solving the above-mentioned problem, as physical tests could be avoided, which are rather costly and time-consuming. But the existing simulation tools do not cover fully the industrial needs, as these either require a lot of computational time or either too complex to use.
In this diploma thesis, a data-driven simulation tool for the laser welding process has been developed, and its main goal is to cover the industrial need, which is the accurate, fast, and simple prediction of the process’s behavior, during a reconfiguration scenario. First, the existing solutions have been examined and analyzed, and the gap between those solutions has been identified. Therefore, the simulation of the process has been carried out in two commercial finite element analysis software tools and used to produce process-related data. Then, through machine learning these data have been used, to create two reduced-order models of the process. The first model is responsible for predicting the distortions of T-shaped geometrical features of a product, while the second model is responsible for predicting the weld bead geometrical characteristics achieved during welding, based on the selected process parameters, and the product’s geometrical features and specifications. In parallel, per the model, the respective way of adoption and use by industry has been developed, based on the industry’s needs.
|
|---|
