Faults are complex structures that substantially influence the mechanical behavior and hydraulic connectivity of rock formations. Therefore, studying faults is important for a variety of disciplines such as geoscience, civil, geotechnical, reservoir engineering, and material science among others. Researchers from these disciplines have considered different aspects of faults, namely geometry, petrophysical properties and mechanics. Until now, these studies have evolved separately and at different scales, making it difficult to connect the geometric development of fault structure to its mechanics. The current understanding of fault geometry and growth is based on fracture mechanics and on many qualitative and quantitative studies on outcrop and seismic reflection surveys among other datasets. The application of fracture mechanics theory is mostly confined to simple geometries: elliptical models for a single fault plane and uniform properties. These applications predict the maximum displacement at the center of the fault, which is not in agreement with the new findings from 3D seismic and outcrop studies. These fracture mechanics models emphasize fault propagation along strike (in 2D). Although they can include the presence of a process zone at the fault tip, the models fail to explain the development of cross-fault damage zones and localization within the fault core as well as fault segmentation and displacement partitioning. Therefore, it is timely to revise the existing applications of fracture mechanics to simple fault geometries and to develop a data-driven fault mechanics possessing closer agreement with real, observed subsurface heterogeneity. This would allow better prediction of fault geometry, propagation, and growth in 3D. We suggest recent advances in non-destructive numerical characterization of faults and application of Deep Neural Networks (DNN) to map fault geometry and predict its properties from seismic data enable us for the first time to extract simultaneously faults' geometrical and mechanical properties at an unprecedented speed and accuracy, thus resolving the 3D fault shape and properties in ways that were unthinkable just a decade ago.

断层是一种复杂的结构，可显着影响岩层的力学行为和水力连通性。因此，研究断层对于地球科学、土木、岩土工程、水库工程和材料科学等多个学科都很重要。来自这些学科的研究人员考虑了断层的不同方面，即几何形状、岩石物理特性和力学。到目前为止，这些研究都是在不同的尺度上独立发展的，因此很难将断层结构的几何发展与其力学联系起来。目前对断层几何形状和生长的理解是基于断裂力学和许多关于露头和地震反射测量以及其他数据集的定性和定量研究。断裂力学理论的应用主要局限于简单的几何形状：单一断层面和均匀特性的椭圆模型。这些应用预测了断层中心的最大位移，这与 3D 地震和露头研究的新发现不一致。这些断裂力学模型强调断层沿走向传播（二维）。尽管它们可以包括断层尖端处过程带的存在，但这些模型无法解释断层核内跨断层破坏带的发展和定位，以及断层分割和位移分区。因此，及时修改断裂力学在简单断层几何形状上的现有应用，并开发一种与实际更接近的数据驱动断层力学，观察到的地下异质性。这将允许更好地预测断层几何形状、传播和 3D 增长。我们建议最近在断层的非破坏性数值表征和应用深度神经网络 (DNN) 绘制断层几何形状并根据地震数据预测其特性方面取得的进展使我们能够首次以前所未有的速度同时提取断层的几何和力学特性速度和准确性，从而以十年前不可想象的方式解决 3D 断层形状和特性。