Calibration of Continuum Cyclic Constitutive Models for Structural Steel Using Particle Swarm Optimization

This paper presents a practical and validated method to automate the calibration of cyclic constitutive model parameters for structural steel. Due to the difficulty in developing homogenous stress-strain fields under large (>10-15%) cyclic strains in experimental specimens, constitutive model parameters are usually calibrated by a trial-and-error approach, wherein the parameters in a finite-element simulation are varied until the simulated response matches the measured response from experiments. This computationally intensive process requires manual intervention and judgment to improve parameter estimates, and can result in nonunique estimates of the model parameters that depend strongly on the calibration specimens. To address this, the calibration exercise is posed as an optimization problem whose objective is to minimize error between measured test data and finite-element simulations of the load-displacement curves for a set of calibration tests. The proposed approach has three elements: (1)development of a robust error measure (objective function); (2)selection of an effective solution search algorithm (the particle swarm algorithm is found to work well); and (3)recommendation of tuning parameters and other best practices for effective performance. The problem is characterized by the high-dimensional search space of a strain-hardening constitutive model, and the computational effort and granularity of the objective function based on continuum finite-element analysis. The methodology is demonstrated to calibrate a widely used cyclic constitutive model from the literature to a series of nine round notched coupon specimens of mild steel. The result is a model fit superior to those achieved through manual calibration with reduced human effort. (C) 2017 American Society of Civil Engineers.