Damage Detection of Composite Materials Using Data Fusion With Deep Neural Networks
Composite materials have enormous applications in the manufacturing of components in field of electronics, spacecraft, and wind energy industry. Thus, it is important to have an efficient damage prediction and prognostics method to avoid catastrophic failures of such components. Since the damage propagation in composite materials is not straightforward, predicting the remaining useful life of composite components is a difficult task. The different damage measurements are often available in different formats and it is very important to consider all of them to achieve a comprehensive analysis for prognostics. Due to this, it is important to employ efficient techniques to consider all the types of damage. Deep neural networks were seen to exhibit the ability to address similar complex problems. The research question in this work is ‘Can deep neural network techniques provide efficient methods to improve damage prognostics of composite materials using data measured from multiple sources?’ To answer this question, two different experimental measurements were taken from NASA Ames Prognostic Repository for Carbon Fiber Reinforced polymer. The specific aims developed were: (1) to perform data fusion of piezoelectric signals and x-ray images and (2) to determine the remaining useful life of composite materials using deep neural network techniques. In order to use data fusion, the acousto-ultrasonic measurements were converted into images using three different image encoding techniques namely, spectrograph, Gramian Angular Field, and Markov Transition Field. The data was then combined with x-ray images taken at specific intervals using data level fusion techniques. The fused data was used as input for the convolutional neural network in combination with Bayesian approach to determine the remaining useful life of composite specimen. The accuracies of all the image encoding algorithms were compared. The analysis showed that using two types of data provides better results than using only one set as it contains information of all the damages that occur in composite materials. Additionally, out of the three image encoding algorithms, Gramian Angular Field was shown to perform the best. The overall determination of the remaining useful life calculation was improved in terms of accuracy and computational expense. Thus, it can be seen that the combination of data fusion and deep neural network techniques provides an efficient method for damage prognostics of composite materials.
Damage Detection of Composite Materials Using Data Fusion With Deep Neural Networks
Category
Technical Paper Publication
Description
Session: 21-05 Data Based Life Prediction
ASME Paper Number: GT2020-15097
Start Time: September 22, 2020, 09:00 AM
Presenting Author: Shweta Dabetwar
Authors: Shweta Dabetwar Texas Tech University
Stephen Ekwaro-Osire Texas Tech Univ
Joao Dias Shippensburg University of Pennsylvania