Development of effective numerical schemes for frictional heat generation and diffusion in vibrothermography

Abstract

Vibrothermography is a non-destructive testing (NDT) method that utilizes high frequency acoustic vibrations to cause flaw faces to rub together and generate frictional heat which is then detectable using an IR camera. This testing process was first investigated over 20 years ago, however it has been slow to develop due to the lack of understanding of the mechanical processes behind the heat generation and the inability to effectively control flaw heating. Vibrothermography is particularly suitable for composite materials due to their tendency to develop subsurface delaminations between laminating plies, which are undetectable using most other NDT methods. Increased use of composite materials in structures such as airplane components and wind turbine blades has contributed to a revived interest in vibrothermography. This paper investigates the use of the finite element analysis (FEA) method to model vibrothermographic systems. Physical samples of glass epoxy composites that contained large delaminations were tested using vibrothermography to create an empirical data set that is used to verify an FE model of a similar system The resulting surface temperatures in the FE model were compared to those observed in the physical test. Both the strengths and shortcomings of using FEA to model these systems are discussed and the proposals for how to improve the model accuracy are provided. The second of half of the paper describes the use FEA to create thermal models of vibrothermographic systems in order to generate a detection model that will indicate the excitation time required to detect a flaw according to it's size, depth, and heat generation rate. The model also accounts for measurement noise and camera distance, and is found to be accurate for flaws with short detection times, and less reliable for those with longer times. The connection between model accuracy and detection time is explained by the inherent issue of problem conditioning. Possible resolutions to this problem are described and further work is proposed on how to improve model accuracy.