Use of Machine Learning in Determining the Parameters of Viscoplastic Models
Please login to view abstract download link
Recently, the machine learning methods proved to be capable of calibrating the parameters of phenomenological constitutive models. In this work, an Arti cial Neural Network based model, Fig.1, is used to directly obtain the parameters of the constitutive models, or to obtain usable initial set of these parameters for further optimization. The inputs of the proposed network are multiple time-series of relevant experiments, e.g. time, stress and strain waveforms. For each experiment the corresponding time-series is processed by a combination of Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) to extract the underlying characteristics of a given sequence. These characteristics extracted from all experiments are then concatenated, and used as inputs for a Feed-forward Neural Network (FNN), which nally computes the constitutive parameters. The performance of the proposed ANN was demonstrated by predicting parameters of a viscoplastic constitutive model with non-linear kinematic hardening and isotropic hardening [1].
complas_sec@cimne.upc.edu/ Telf. + 34 - 93 405 46 94
Copyright © 2022 CIMNE, All Rights Reserved. Terms of service
