SYMPOSIA PAPER Published: 01 January 1998

Residual Strength Prediction of Impact-Damaged Composite Structures by Optical and Acoustical Computer Sensing with Neural Network Techniques


The proliferation of composites as a structural material has increased the necessity for detection and characterization of impact damage. Impact damage, due either to mishandling or in-service environment, in the form of delaminations, matrix cracking, and fiber chopping, may be sufficient to reduce the load-bearing ability of the structure below safety limits and may still not be visible to the unaided observer. Such damage may exist on the surface of an internal cavity, which may prevent it from being viewed, or it may exist within the thickness of the structure and not be visible from either side of the laminate. Impact damage that is partially visible often extends further along and through the material than that which can be seen. Computerized methods for damage detection offer greater assessment capabilities than visual inspections and are capable of detecting subsurface or internal flaws. For these reasons, as well as the declining cost of more powerful hardware, computer sensing methods for nondestructive evaluation (NDE) have grown increasingly popular in recent years.

A disadvantage of some computerized NDE methods is that vast amounts of data are generally obtained from any given test, and often it is up to the inspection conductor to interpret these results subjectively based upon prior experience. An alternative is to train the computer sensing system to analyze the huge data arrays, which it can digest more easily than can its human counterpart, and to provide objective interpretations. This paper presents the results of an experimental study in which two computer sensing techniques were used to monitor filament-wound pressure vessels during pressurization. Acoustic emission registers the sound generated by microscopic damage propagation. Video image correlation is a noncontact computer vision technique that simultaneously measures full-field in-plane surface displacements and strains, both linear and angular, with subpixel accuracy. Neural networks were used to predict the burst pressures of impacted pressure vessels based upon data obtained at less than approximately one third of the expected burst pressure for an undamaged specimen.

Author Information

Lansing, MD
University of Alabama in Huntsville-Research Institute, Huntsville, AL
Walker, JL
University of Alabama in Huntsville-Center for Automation and Robotics, Huntsville, AL
Russell, SS
NASA Marshall Space Flight Center, MSFC, AL
Price: $25.00
Contact Sales
Reprints and Permissions
Reprints and copyright permissions can be requested through the
Copyright Clearance Center
Developed by Committee: E08
Pages: 285–297
DOI: 10.1520/STP13279S
ISBN-EB: 978-0-8031-5382-0
ISBN-13: 978-0-8031-2609-1