An ANN based approach for the friction stir welding process intrinsic uncertainty

Friction Stir Welding is a solid-state bonding process that during last years has caught the researcher's attention for the mechanical characteristics of the welded joints that are quite similar to the properties of the base material. The Friction Stir Welding is affected by several process parameters leading to intrinsic variability in the process. The present paper would introduce a new approach for predicting the surface hardness in different areas of the welded parts. Specifically, this method is based on the hypothesis that multiple Artificial Neural Networks, characterized by the same architecture but different weights, can be used for forecasting both the punctual value of the local hardness and its confidence interval, resulting in taking into account the intrinsic variability of the process.