Ramin Saberi

Grade:  Master

Thesis Title:

Prediction of ductile damage based on a modified Lemaitre damage model considering the Lode parameter using artificial neural networks

Year: Sept. 2020 - Oct. 2023.

Abstract:

In this research, a computational method using artificial neural networks to predict ductile damage based on the extended Lemaitre damage model. Damage prediction is usually performed using the finite element method and includes various parts such as modeling the component geometry, various loads under different working conditions, selecting the appropriate damage model, creating a finite element model, and then implementing and extracting the results. This process must be repeated as the working conditions of the workpiece change and it is necessary to continuously monitor the damage parameters, which can be a time-consuming and costly process. With the increasing processing power of computer systems and the high speed of data-driven methods, the desire for data-driven damage prediction mechanisms is growing today. In this study, the data-driven artificial neural network method was presented as a different approach from the finite element method for predicting the stress tensor and damage. Continuous damage models have attracted much attention in recent years and much research has been conducted in this area. In this study, the improved Lemaitre damage model with Lode parameters was used to calculate the damage. The use of data-driven methods in complex phenomena requires access to a considerable amount of data in order to properly train the network and achieve its optimal performance. The calculation of the stress tensor and the damage in the material depends on the loading history of the material and is a function of the total strain and plastic strain values. Therefore, to reduce the number of simulations and obtain more data, the stress tensor and damage data of different parts of the workpiece were used in addition to one point of the workpiece in each analysis. The Lemaitre damage model was implemented together with the Lode parameter in a user material subroutine (UMAT) in the Abaqus software. A further subroutine was then developed to automate the analysis in the Abaqus software. The analysis was performed for 400 samples of loading data, and the values of total strain, plastic strain, stress tensor and damage amount were extracted. Next, the network structure, i.e. the number of layers, the number of neurons and the corresponding model of the problem were introduced. Recurrent neural networks (gated recurrent units) have a memory and can memorise parts of the course of the network's input parameters. The parameters of the neural network's learning algorithm are determined by trial and error in order to achieve better performance. To assess the capability and accuracy of the neural network, some of the data that was not used to train the network was used to evaluate and benchmark the network. The error plots obtained from the results of the artificial neural network model and the results of the finite element model show the accurate performance and realistic prediction of the artificial neural networks.

Keywords: Artificial neural networks, Enhanced Lemaitre damage model, Lode parameter, Deep learning, Gated recurrent unit.