Unsupervised and supervised data-driven techniques to characterize laboratory-scale mechanical discontinuity

Abstract:

Active ultrasonic measurement was integrated with passive acoustic-emission measurements using unsupervised learning and data fusion techniques to leverage the advantages of the two modalities of measurements. Unsupervised clustering methods, namely DBSCAN, agglomerative, and K-means, processed the laboratory-based ultrasonic shear-waveform measurements and acoustic emission waveforms on Tennessee sandstone after hydraulic fracturing. Based on displacement discontinuity theory, each cluster label determined by the clustering method is then associated with a specific degree of geomechanical alteration (change of stiffness) in the material due to hydraulic fracturing. Use of short-time Fourier transform followed by robust scaling and principal component analysis ensures that various clustering methods generate relatively similar clustering labels. Two sets of two-dimensional maps of fracture and fracture-induced damage in axial, median and frontal planes were obtained by separately processing the active and passive measurements using unsupervised learning methods. These 2D maps are then fed to wavelet-based image-fusion technique to integrate the two sources of information to reliably image the embedded fractures and the surrounding geomechanically altered regions.

Wave propagation and diffusive transport interacts with mechanical discontinuities (i.e. cracks and fractures). Data-driven classifiers were developed on simulated dataset of multipoint compressional, shear, and pressure wavefront travel times with associated labels categorically describing the static fracture network embedded in 2D material. Such classifiers can categorize the fractured materials using only the multipoint wavefront travel times. The data-driven classifiers can not effectively categorize fracture networks of different dispersions around one primary fracture orientation. Voting-based ensemble classifier has accuracy of 0.99 when categorizing fractured materials having primary fracture orientations differing by 45° despite fracture dispersion of +/-20° around the primary fracture orientation. Materials with random, unimodal, bimodal, or linear fracture distribution can be categorized at a high accuracy of 0.86.