Title:
Synergizing Machine Learning and Nonlinear Finite Element Analysis to Simulate the Behavior of UHPC Members
Author(s):
Diab
Publication:
Web Session
Volume:
ws_S24_Diab.pdf
Issue:
Appears on pages(s):
Keywords:
DOI:
Date:
3/23/2024
Abstract:
This study delves into the feasibility of synergizing machine learning-driven constitutive models with nonlinear finite element analysis (NLFEA) to simulate the structural response of UHPC members. One of the major challenges in predicting the response of UHPC members consists of characterizing the UHPC tensile response. This is primarily due to the multitude of factors that influence it, such as parameters related to the mix design, type of fibers used, and the type of test performed. As such, machine learning has the potential to become a useful tool in predicting the behavior of UHPC in tension due to its ability to map the convolutional relationships between multiple input parameters. Consequently, the applicability of three distinct machine leaning models to predict the UHPC tensile response was investigated. Three distinct machine learning models were used to predict the tensile response of the UHPC material, including artificial neural networks (ANN), decision trees, and evolutionary algorithms. Principal component analysis was conducted to discern the main parameters influencing the tensile behavior of UHPC. Among the machine learning models investigated, the ANN displayed the most accurate predictions. The ANN-generated UHPC tensile properties were subsequently employed within a NLFEA methodology that uses a secant-stiffness, smeared, rotating-crack formulation to simulate the response of UHPC members tested in the literature, including panel elements subjected to pure shear, shear-critical and flexural-critical beam specimens. The numerical results yielded a reasonable level of accuracy in simulating the behavior of the elements in terms of stiffness, strength capacity, crack pattern and failure mode.