Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
A machine learning approach for the fast simulation of fatigue-induced crack propagation
Aim of this project is to develop a surrogate model based on machine learning approaches that allows a fast yet reliable numerical simulation of the fatigue-induced cracking process in mechanical components subjected to a high number of cycles.
The prediction of the response of a structural component when subjected to fatigue loading is of key importance in the civil, aerospace and mechanical industry. This is classically done using empirical rules that, however, lack of generality, namely a new formula must be calibrated for each component geometry and/or loading condition. The adoption of the finite elements along with fatigue constitutive laws is a more flexible approach that, on the other hand, might require an unbearable computational effort when simulating directly and cycle-by-cycle the material response of a component subjected to a large number of cycles. This calls for fast alternative approaches able to drastically reduce the computational effort while preserving flexibility and accuracy of the solution.
The main aim of this project is to develop a surrogate model based on machine learning that allows the numerical simulation of the fatigue induced cracking in mechanical component subjected to a high number of cycles. The surrogate model will be trained using digital data obtained from an in-house code and then implemented in a finite element code allowing for a fast yet accurate prediction of the structural fatigue response.
The prediction of the response of a structural component when subjected to fatigue loading is of key importance in the civil, aerospace and mechanical industry. This is classically done using empirical rules that, however, lack of generality, namely a new formula must be calibrated for each component geometry and/or loading condition. The adoption of the finite elements along with fatigue constitutive laws is a more flexible approach that, on the other hand, might require an unbearable computational effort when simulating directly and cycle-by-cycle the material response of a component subjected to a large number of cycles. This calls for fast alternative approaches able to drastically reduce the computational effort while preserving flexibility and accuracy of the solution.
The main aim of this project is to develop a surrogate model based on machine learning that allows the numerical simulation of the fatigue induced cracking in mechanical component subjected to a high number of cycles. The surrogate model will be trained using digital data obtained from an in-house code and then implemented in a finite element code allowing for a fast yet accurate prediction of the structural fatigue response.
1. Literature study on fatigue mechanics and machine learning approaches
2. Generation of the training dataset
3. Formulation, implementation and validation of a machine learning algorithm
4. Comparison between high-fidelity and surrogate predictions and critical discussion
After implementing the main tasks, the project may move on in multiple possible directions, giving room for inputs from the student's side.
1. Literature study on fatigue mechanics and machine learning approaches 2. Generation of the training dataset 3. Formulation, implementation and validation of a machine learning algorithm 4. Comparison between high-fidelity and surrogate predictions and critical discussion After implementing the main tasks, the project may move on in multiple possible directions, giving room for inputs from the student's side.
Dr. Pietro Carrara
Computational Mechanics Group
Tannenstrasse 3, CLA J 17.1
E-mail: pcarrara@ethz.ch
Dr. Pietro Carrara Computational Mechanics Group Tannenstrasse 3, CLA J 17.1 E-mail: pcarrara@ethz.ch