This research aims at determining quantitatively the mechanical response state of an engineering material, as based on quantitative non-destructive testing combined with “pattern recognition and classification” methodologies. In the pertaining experimental procedure, stress waves are “simulated” in the microstructure of the material to resemble acoustic emission waves but without disrupting the material, i.e., without a simultaneous application of an external loading. After propagating through the microstructure, the waveforms are captured, identified, and then classified as belonging to various classes, where each class represents one of different states of the tested material-property. The presented approach has been proven to be powerful in determining quantitatively numerous material response states in both homogeneous and heterogeneous classes of engineering materials that were subjected priori to unknown static, quasi-static and dynamic types of loading. The presented study illustrates meantime the influence of the “normalization procedure” on the testing of the computer-based “Classifier” performance, whereby a number of “Normalization Trees” are built for each multiple-class problem. In this, the experimental results show that, for the same PR-classifier, when identifying the same material mechanical state, significant differences of classification performance may be obtained by employing different normalization trees.

The presented approach has the strong potential of determining non-destructively the ultimate strength of the material in a quantitative manner.

KEYWORDS: Non-destructive inspection, Response of materials, Pattern recognition