Structural Health Monitoring using the Finite Element Method and Deep Learning

Abstract: The work develops a Data Driven pipeline for monitoring the structural health of one-dimensional slender structures using sparse sensor measurements. Two distinct problems are addressed: the static and dynamic response of an Euler-Bernoulli beam. A linear regression model is trained for the static problem while a Recurrent Neural Network with Long Short Term Memory units is used for the dynamic one. The generation of the training dataset is obtained by means of a finite element code implemented ad-hoc in Python while the training itself is performed using Tensorflow / Keras and scikit-learn. Some results are presented to validate the proposed method and to evaluate the predictive capabilities as the type of data available varies (e.g. different positioning of the sensors).

slide2

The results were presented at the ESA/NASA 2022 International Conference of Advanced Manufacturing

Keywords: Deep Learning, Machine Learning, scikit-learn, Google Cloud Platform, Keras/Tensorflow, Python, Finite Elment Method.