Mapping discrete fracture networks using inversion of hydraulic tomography data with convolutional neural network: SegNet-Fracture

• Mapping the fracture network from hydraulic heads in a fractured aquifer. Directly approximate inverse function using neural network. Principle based on an encoder-decoder convolutional An end-to-end operator with accurate mapping, perform instantly. quality relies nature of dataset, resists to data noise. In this paper, we propose new method map structure heterogeneous aquifer inversion head measured during pumping tests tomography mode. This tool is concept networks, which provides direct approximation linking geometry data. order handle highly nonlinear more effectively, advanced developed SegNet architecture structure, excels image processing translate water level associated at input into output. The trained synthetic dataset where and matrix heterogeneity are randomly generated, obtained by solving groundwater flow equation. accurately maps different complexity levels fractures embedded transmissivity. As data-driven approach, accuracy mapping depends quantity, quality, relevance used training phase. While generating train requires effort, performs each result appears be stable even presence noise, reliably interprets if they share comparable properties as specified models.