Numerically Efficient Fatigue Life Prediction of Rocket Combustion Chambers using Artificial Neural Networks

Abstract

Fatigue life prediction is an essential part of multidisciplinary design studies and optimization loops, but state of the art finite element based methods are numerically inefficient. We overcome this challenge by training an artificial neural network to predict the number of cycles to failure, based on combustion chamber geometry and operational point. To accomplish this, a 2-d finite element analysis generates 250 000 training data samples. The trained network then predicts previously unseen data with a mean absolute percentage error of 6:8 % in less than 0:1 ms per sample compared to up to 5 min with finite element based methods. To the best of our knowledge, this publication is the first to successfully apply machine learning to fatigue life prediction.