Analytical models’ driven machine learning modeling for the setting and control of the radial-axial ring rolling process

Radial axial ring rolling (RARR) forming is focal in seamless ring production for several industrial sectors. Although industrial settings and decisions are still largely driven by operators’ experience, this work provides a real-time and interactive solution for parameters’ settings and control. In fact, the high computational time required for a simulation together with the required knowledge to set a finite element model (FEM) disadvantages this solution when a fast and reliable decision is required. The proposed approach combines analytical and gradient boosting regression models based on experimental and simulated data, to develop a software that offers a real-time and interactive simulation for the production of flat rings. The algorithm predicts instantaneously the manufacturing forces, torques, and power required, starting from the final ring geometry, tools speeds, and machine data. The proposed solution has been verified with industrial case experimental data and with the replicated simulation on Simufact Forming 15 FEM software and predictions show an average which is above 90% on different process settings and ring materials and dimensions.