A framework for calibration of self-piercing riveting process simulation model

This paper presents a framework that integrates machine learning and global sensitivity analysis to calibrate process simulation model of self-piercing rivet (SPR). In the present framework, surrogate models are trained to represent the intrinsic numerical relationship between selected model parameters (e.g., material properties, interface frictions, and clamping force) and cross-section dimensions of SPR joint obtained from simulation. Under the assistance of surrogate models, a global sensitivity analysis is performed to identify critical parameters that have significant effects on SPR joining process. Then, new surrogate models are trained to present the sensitive parameters and the cross-section dimensions obtained from process simulation. The parameter calibration of the simulation model is described as a two-objective optimization problem to minimize errors of key cross-section dimensions between simulation and experiment. Combined the newly trained surrogate models with genetic algorithm, the model parameters are calibrated efficiently. A case study of simultaneous calibration of four SPR process simulation models is used to illustrate the proposed framework and examine the effectiveness of the proposed framework. A good consistency is achieved between calibrated models and experiments.