Post peel test assessment of metal-metal bonded interface using hyperspectral imaging

Abstract

The adhesion process of bonded structures is a very sensitive manufacturing process, namely to surface contaminations. Nowadays, the adhesion quality of bonded joints can only be detected by destructive testing such as peel tests. But, for actual structures destructive testing of every bonded part is obviously not an option and industry is lacking a non-destructive test to quantify adhesion strength. In this work a methodology is described that could lead to a successful detection of possible contamination agents during the manufacturing process of bonded structures. Hyperspectral imaging is an optical non-destructive technique that allows mapping of specific chemical characteristics of a surface in high contrast. It is a well-recognized technique that has already been successfully applied in a wide range of fields including astronomy, remote sensing, cultural heritage and medical sciences.
Multiple sets of aged and non-aged peel tests were destructively tested under different conditions. One type of adhesively bonded specimens was studied: aluminium-to-aluminium using standard floating roller peel test (ASTM D3167). Five sets of samples with different aging conditions were tested. After destructive testing, the contamination level of the fracture surfaces of the bonded specimens were studied using hyperspectral imaging. The hyperspectral imaging system used was a line-scan spectral imaging device (Imspector V10E spectrograph, SPECIM, Finland) capable of acquiring data in the range of 400 nm – 1000 nm with a bandwidth of 2.8 nm. Three tungsten halogen light sources (30W, Osram) were used for uniform illumination across all the sensitivity range of the sensor. The resulting images of each specimen had a resolution of 1312 * 300 pixels, with an exposure time of 200 msec. Following acquisition, all measurements were normalized with the use of a white diffuse reference target (WS1, Ocean Optics). Reference reflectance spectra were extracted from suspicious areas and a non-negative-least-square un-mixing algorithm was applied to check for chemical similarities. Two main signals were detected and mapped onto the surfaces. Statistical analysis was performed, based on the mapping results that described the surface percentage coverage of the contaminant signals and the related statistical errors. A correlation between the percentage of contamination and the adhesive failure was found. Results are promising and will be further studied with other techniques and chemical analyses to fully understand the level of contamination.