Sviluppo di tecniche produttive per l’ottimizzazione integrata di materiale, geometria e processo basata su sistemi di apprendimento automatico

In recent years, machine learning has become an important innovative tool for improving manufacturing processes. It can find patterns in data, give quick predictions, and help control processes more effectively. Traditional methods like analytical models and finite element method (FEM) simulations are reliable but have limits. Analytical models are often too simple, while FEM simulations are accurate but slow and expensive. Machine learning offers a faster option that can still give reliable results and support real-time decisions. This research applies machine learning to two well-known manufacturing processes. The first is short fiber reinforced thermoplastic injection molding. In this process, fiber orientation strongly affects the final product and numerical simulations to predict fiber orientation require computational time and cost. A machine learning model was created to quickly predict fiber orientation based on geometry, gate position, and process settings. This makes it easier to test different designs and improve the material performances. The second case is hot radial-axial ring rolling, used to make large metal rings. Predicting forming forces and torques is difficult with traditional analytical formulations and require time with FEM simulations. A hybrid model combining machine learning and analytical tools was trained with industrial data to predict forces, torques and energy consumption in real time.