For several years, SAFRAN use a descriptive numerical model of the inertial friction welding process. This model provides a better understanding of physical phenomena during the process (in particular, temperatures assessment in both workpiece). This kind of model has some limitations in terms of heat transfer at the interface between the two parts. It requires input data from welds already carried out. Therefore, it cannot be used as a predictive model. In order to tackle this limitation, SAFRAN identified a Stribeck friction curve thanks to a neural network. It allows to predict during the welding simulation the heat transfer at the interface between the two parts. The model needs geometric data (mean diameter and thickness) and process parameters (surface energy, pressure, linear speed) as input. Identifying this model required a large experimental database with several hundred welds on different materials and specimen sizes. The inertial friction welding machines available within the SAFRAN group have been used in order to build this database. The neural network model has been trained on 90% of the experimental database. The remaining 10% was dedicated to the model validation by correctly identifying the friction law at the interface. Once this neural network set up, its validation was performed by running numerical simulations and validating the thermal gradient within the part, the spindle deceleration rate, and the numerical material consumption at the end of the process. This methodology is the subject of a patent application, reference number: B-026789FR