A data-driven approached is developed to extract shear and compressional slowness from dispersive sonic logging waveforms, or waveforms in which the formation arrival is interfered by tool arrival of similar but different slowness. It relies on extracting the dispersion curve, or slownessfrequency coherence map, from waveforms for a range of frequencies and slownesses. The dispersion curve is then converted into an objective function such that the true formation slowness is associated with an anomaly of the function. Specifically, the objective function can be a probability density function or a histrogram of slownesses extracted from the dispersion curve. The formation slowness coincides with one of the local maxima of the histogram, and can be automatically determined with an optimization algorithm. Alternatively, the objective function can also be derived by summing the slownessfrequency map across the selected frequency range. The formation slowness is then determined by finding the local maximum or minimum of the nth derivative of the objective function, depending upon the character of the data. In either case, the objective function is plotted as a color map for a range of slownesses and depths. This color map is a better QC tool than the widely used semblancebased coherence map, as it reflects the true slowness populations in the waveform. The method has been extensively tested and validated using synthetic data, wireline and LWD sonic data from 20 wells across six geographically-diverse regions.
Formation compressional and shear slowness are two of the most important parameters used in the exploration and production of hydrocarbon. Conventionally, they are measured by sonic logging. The acoustic source on the logging tool sends signal that subsequently propagates along the well and is recorded at several evenly spaced receivers that are some distance away from the source. Formation compressional and shear slowness are then estimated by processing the recorded waveforms, using array sonic processing techniques, such as the popular slowness-time coherence (STC) method (Kimball and Marzetta, 1986). Recent studies have shown that STC only yields accurate slowness estimation when and only when the waves propagating along a wellbore are non-dispersive, or multiple arrivals contained in the waveforms are well separated in the slowness-time domain. When the underlying waveforms are dispersive or the waveforms compose of mixed modes with similar slownesses, in such cases as the leaky P-mode, dipole mode, quadrupole mode or even compressional mode in LWD, STC produces a systematic error in the slowness estimation (Kimball, 1998; Geerits and Tang, 2003; Goldberg et al, 2003; Valero et al, 2004). The amount of those systematic errors in many cases, such as the case dipole logging in oil-based mud, is large enough to cause problems in reservoir characterization. And moreover, the correlogram obtained from STC analysis, widely used as a quality control tool for slowness estimation, does not reflect the accuracy of the slowness estimation. Recently, several approaches have been developed to address the described problem. They fall into two categories: model-driven dispersion correction and phase slowness processing.