Fringe patterns recognition in digital photoelasticity images using texture features and multispectral wavelength analysis

In digital photoelasticity, fringe pattern analysis is crucial because the photoelastic fringes provide information about direction and magnitudes of the principal stresses at the surface of the inspected object. These fringes exhibit visual properties that depend on the applied load, their spatial location in the inspected object geometry, and the illumination source. Traditional methods for fringe analysis in photoelasticity have limited performance when dealing with noisy or not well contrasted fringes, or if the spatial resolution of the fringes is lost. This work presents an approach for analyzing fringe patterns in photoelasticity images using texture information, in conjunction with machine learning techniques. Stress fields are simulated in multiple spectral bands for two models. Then, different regions of interest in these models are characterized with well-known texture descriptors. Furthermore, feature ranking and five classification schemes are used to describe the texture variations that occur in the models when they undergo diametral compression in the different spectral bands considered. The results show that texture descriptors are suitable tools for describing the stress information provided by photoelastic fringe patterns. Also, it is possible to use machine learning techniques to learn, recognize, and predict the behavior of models subjected to mechanical load in photoelasticity experiments.