Speaker
Description
Within the context of hot forming of metallic alloys and the optimization of process parameters, an intelligent surrogate model would be of interest for the identification of material behavior using industrial data. An accurate constitutive equation aims to improve the ongoing forging process. Furthermore, online calibration of the physical and/or constitutive parameters of the working material enables process monitoring that integrates artificial intelligence and our knowledge of the mechanics of materials [1]. We expect more interpretable process monitoring. For a preliminary feasibility analysis, we consider a synthetic dataset for training, but both experimental and synthetic data for testing set. The calibration process uses a dataset of strain-stress curve at different temperatures and strain rates as input data. The calibration step's output is the ten coefficients of the Hansel-Spittel flow stress model. We propose comparing the classical optimization process for finding optimal parameters with the model parameters predicted by an intelligent surrogate model. This intelligent surrogate model is a specialized artificial neural network consisting of fully connected layers. In this work, the training objective is to perform a regression task as in [2], we discuss the size of the required training dataset, calibration accuracy, and new opportunities enabled by intelligent surrogate modeling. This discussion covers cases of experimental data that do not conform to the Hansel-Spittel flow model. This results in a calibration error.
| Speaker Company/University | Thanh Chung Nguyen |
|---|