Optimized and tailorable speckle pattern generation approach for digital image correlation applications in SHM

A novel speckle pattern development technique is introduced to create a parametric space to allow for implementation of the Digital Image Correlation (DIC) method in Structural Health Monitoring (SHM) applications. DIC is a non-contact, vision-based deformation measurement method that has been widely used in mechanical testing and material characterization, as well as in structural testing applications. Being scale invariable, this technique has been used to measure full field 3D deformation as well as to identify damage occurring on a large range of scales from transportation bridges all the way down to material microstructure measurements. DIC tracks subsections of images taken during the deformation process to provide full field displacements and maps. To accomplish this goal, surface patterns are used when employing stereoscopic vision and the results with respect to deformation resolution vary due to both hardware and software reasons. In general practice, spraying and airbrushing are used for speckling in a scale from millimeters to meters, while stencils and specialized markers are applied to track deformation on bigger structures, the quality of which is dependent on the user's experience. In this paper an optimized pattern development technique is proposed to numerically and on-demand generate speckle sizing and distribution-controlled speckle patterns to reduce the uncertainty and increase the usefulness of DIC measurements at variant scales. The quality of the pattern is achieved by extracting features from a bio-templating inspired pattern and merging them with size control parameters for optimized pattern generation for given FOV sizes. Features that improve the quality of the speckle pattern are subsequently extracted via a deconvolution within the frequency spectrum. Gradient information controlling properties of the high-quality pattern are then synthesized and convolved with size and distribution controlling properties of a numerically generated pattern to produce optimized patterns which can then be transferred onto the specimens/structures. An error analysis study shows that the optimized pattern provides better results as compared to traditional speckling methods and that of the original bio-templated image. While the systematic bias in the error remained comparable to that of standard speckling methods, the random bias error reflects an 80% reduction across the entire field of view investigated. The viability of producing multiscale deformation information using this novel approach is assessed by a number of numerical tests.