A new interpretation workflow to automatically deliver high-resolution dynamic elastic moduli using borehole sonic data is presented. This workflow consists of three new algorithms: a multi-resolution based depth tracking algorithm to avoid mislabeling among the processed slownesses of different depths; a physics-based machine-learning enabled flexural dispersion extraction algorithm that does not require zone-by-zone tuning of mud slowness, borehole size, and frequency filters; and a new inversion algorithm that jointly inverts the three Thomsen anisotropic parameters as well as mud slowness as a function of depth. Deliverables of this interpretation workflow include rock dynamic anisotropic elastic moduli in a way that is automatic and reliable. This workflow has been applied to several field datasets and can yield more reliable anisotropic shear moduli than the traditional interpretation method. Additionally, the advantages of this fully automated workflow against the traditional model-based core-log integration method are demonstrated through these field data applications.
Geomechanics plays an important role in the exploration and development of oil and gas reservoirs, for which 1D and 3D mechanical earth models (MEM) (Plumb et al., 2000; Berard and Prioul, 2016) are often built as the foundations to further apply advanced solutions, such as borehole stability, sand production, hydraulic fracturing, and reservoir production performance. One of the most essential inputs to build the MEM is the continuous anisotropic elastic moduli (cij) interpreted from borehole acoustic data. These elastic moduli are calculated from the bulk density and sonic slownesses. The sonic slownesses include the compressional, fast, and slow shear logs, as well as anisotropic parameters represented by the three Thomsen parameters. These sonic slownesses are interpreted from monopole, dipole and quadrupole waveforms acquired by a wireline or logging-while-drilling (LWD) borehole acoustic logging tool, illustrated in Fig. 1. After logging, the waveforms are processed using signal processing algorithm. However, existing interpretation workflow often requires many steps, such as sonic log re-labeling, missing data patching, and careful quality control. The complexity of the sonic interpretation workflow presents a significant bottleneck for the timely delivery of 1D and 3D MEM building, especially with large volume of acquired sonic data.