Self-piercing riveting (SPR) is a high-speed cold mechanical fastening technique suitable for joining two or more metal sheets. SPR has become widely used in the automobile industry, due to the increasing use of new lightweight materials (such as aluminium) and alloys, or even dissimilar material combinations that are difficult or impossible to weld. The finite element method (FEM) has proven to be highly efficient for better understanding the mechanisms involved in the SPR process as compared to the experimental procedures which are costly and time consuming. Hence, the experimental trial and error approach could be avoided. However, this approach needs the use of accurate and reliable constitutive law, damage model, friction coefficients (at each contact pair) in order to obtain satisfactory results. Consequently, it becomes costly in terms of data entry and calculation time [1]. The work aims at using a machine learning (ML) technique to evaluate the final cross-sectional features of the riveted joint (hence its quality) from only the punch force-displacement curve. For this purpose, first a 2D-axisymmetric model is developed using a commercial software. Then, a sampling of numerical solutions is used to train a deep-learning [2] model as follows: an autoencoder is employed to reduce the dimensionality of the force-displacement curves to the reduced-dimension latent space. A regression between the latter and the output cross-sectional features is then carried out by using a multilayer perceptron (MLP). REFERENCES [1] He X., Pearson I. and Young K., Self-pierce riveting for sheet materials: State of the art, Journal of Materials Processing Technology, vol.199 (12), pp. 27-36, 2008. [2] Goodfellow I., Bengio Y. and Courville A., Deep Learning. Springer US, 2016.