Self-piercing riveting process: Prediction of joint characteristics through finite element and neural network modeling

Developing robust numerical simulation tools to investigate the self-piercing riveting process is critical, since the feasibility, quality and strength of riveted connections relies on the successful formation of mechanical interlocks between the rivet and sheet materials. In the present study, we investigate the riveting of aluminum alloy and dual-phase steel sheets (AA7075-F/DP600 and AA6016-T4/DP600) over a wide range of sheet thicknesses, as a function of the rivet and die geometries employed. More specifically, we study the dependence of the probability of successful joint formation, defined as the ratio of the number of acceptable riveted connections to the total number of test cases, on the rivet leg inner radius and die central tip depth. Towards this, we use experimentally calibrated Hosford-Coulomb fracture surfaces for each deformable part, incorporated in dedicated axisymmetric 2D finite element (FE) process models. The FE predictions are validated through comparisons with experimentally obtained riveted joints. Moreover, we analyze the relation between the mechanical interlocks achieved and the rivet and die geometries employed, deriving practice-relevant conclusions with respect to the most favorable design parameters. In particular, we show that while the probability of successful joint formation decreases upon increasing rivet leg inner radius, the joint quality, in terms of effectuated interlock distance per rivet mean residual equivalent plastic strain, increases. Furthermore, we show that machine-learning techniques can be employed to classify with remarkable accuracy the successful joint formation for a wide range of possible self-piercing riveting scenarios, accounting for both rivet and die-related geometrical attributes, as well as for the thickness of the metal sheets.