Prediction of the transverse elastic modulus of the unidirectional composites by an artificial neural network with fiber positions and volume fraction
Do-Won Kim, Shin-Mu Park, Jae Hyuk Lim
Abstract
Abstract This paper presents the development of artificial neural network (ANN) modeling for predicting the transverse elastic modulus of a unidirectional composite (E-glass/MY750) with fiber positions and volume fraction. For this prediction, random representative volume elements were generated according to the fiber volume fraction ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mi>V</mml:mi> <mml:mrow> <mml:mtext>f</mml:mtext> </mml:mrow> </mml:msub> </mml:mrow> </mml:math> ). A training dataset consisting of input data ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mi>V</mml:mi> <mml:mrow> <mml:mtext>f</mml:mtext> </mml:mrow> </mml:msub> </mml:mrow> </mml:math> and fiber locations) and output data (the effective elastic modulus), which were computed by the computational homogenization scheme (CHS), was used to train the ANN model to have proper weight and bias by back propagation. To demonstrate the performance of the proposed ANN model, prediction of the transverse elastic modulus of various test datasets whose transverse elastic moduli are known by CHS was conducted. The prediction accuracy was verified in terms of the mean squared error, correlation coefficient ( R ), and relative error. The prediction results showed excellent agreement with the test dataset and quickly predicted the transverse elastic modulus having random microstructures.