Neural network based nondestructive evaluation
of sandwich composites
Frederick Just-Agosto
*
, David Serrano, Basir Shafiq, Andres Cecchini
College of Engineering, University of Puerto Rico, Mayagu
¨
ez, Box 9044, PR 00681, USA
Available online 12 March 2007
Abstract
A neural network using a combination of complementary vibration and thermal damage detection signatures is proposed. Sandwich
composites consisting of two carbon fiber/epoxy matrix face sheets laminated onto a urethane foam core were experimentally and analyt-
ically characterized for vibration, and thermal response. The numerical models developed were later used to establish neural network train-
ing data. Results show that the network can successfully detect damage when using just a single method vibration or thermography fails.
Ó 2007 Elsevier Ltd. All rights reserved.
Keywords: D. Non-destructive testing; C. Finite element analysis
1. Introduction
Sandwich structures have extremely high flexural stiff-
ness to weight ratios that offer alternatives quite attractive
to designers. Examples of use can easily be found in the
marine, aerospace and civil industry. The disadvantage,
however, is that composite materials fail in complex failure
modes that are difficult to detect [1]. No single NDE tech-
nique is physically capable of detecting all damages and the
selection of appropriate NDE techniques is, therefore, a
very important issue in order to improve service life.
Any localized defect in a structure produces variations
in the dynamic response. Although many methods are
available, several authors have successfully examined the
effects of curvature changes to detect damage in structures
[1–4]. On the other hand, stimulated infrared (IR) thermog-
raphy is used for the localization and characterization of
thermally resistive defects. This NDE method is gaining
popularity in finding defects in engineering structures due
to its robustness and easy implementation [5–8].
While there are many approaches that have been inves-
tigated and are still being developed for damage signature
based NDE, improvements can be obtained by using neu-
ral networks (NN) to properly discern various mutually
independent dynamic system anomalies and structural
defects [9–12]. The present work proposes to combine
vibration and thermal signatures to locate and quantify
damages in foam core sandwich composite structures using
NN scheme. This approach is significant as it combines the
two NDE techniques instead of just using one as generally
performed [9–13].
2. Nondestructive evaluation
The response to different types of damages, such as, face
sheet debond, small perforations, and cracks in a cantilever
beam were investigated using mode shape curvature and
transient temperature response. The sandwich composite
beams studied consisted of bi-directional woven [0°/90°]
carbon fiber face sheets bonded to polyurethane foam
core with epoxy resin. Nominal beam dimensions were
584 mm · 51 mm with face sheet thickness of 0.7 mm and
core thickness of 6.4 mm. Finite element and finite volume
approaches were employed for numerical simulation. Sev-
eral damage scenarios were experimentally tested with both
techniques to validate the numerical simulation. Finally,
virtual neural network training was performed based on
data obtained from numerical models.
1359-8368/$ - see front matter Ó 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compositesb.2007.02.023
*
Corresponding author. Tel.: +1 787 832 4040x2546; fax: +1 7878 265
3817.
E-mail address: fjust@me.uprm.edu (F. Just-Agosto).
www.elsevier.com/locate/compositesb
Available online at www.sciencedirect.com
Composites: Part B 39 (2008) 217–225