No abstract

Petrophysics is a pivotal discipline that bridges engineering and geosciences for reservoir characterization and development. New sensor technologies have enabled real-time streaming of large-volume, multi-scale, and high-dimensional petrophysical data into our databases. Petrophysical data types are extremely diverse, and include numeric curves, arrays, waveforms, images, maps, 3-D volumes, and texts. All data can be indexed with depth (continuous or discrete) or time. Petrophysical data exhibits all the "7V" characteristics of big data, i.e., volume, velocity, variety, variability, veracity, visualization, and value. This paper will give an overview of both theories and applications of machine learning methods as applicable to petrophysical big data analysis. Recent publications indicate that petrophysical data-driven analytics (PDDA) has been emerging as an active sub-discipline of petrophysics. Field examples from the petrophysics literature will be used to illustrate the advantages of machine learning in the following technical areas: (1) Geological facies classification or petrophysical rock typing; (2) Seismic rock properties or rock physics modeling; (3) Petrophysical/geochemical/geomechanical properties prediction; (3) Fast physical modeling of logging tools; (4) Well and reservoir surveillance; (6) Automated data quality control; (7) Pseudo data generation; and (8) Logging or coring operation guidance. The paper will also review the major challenges that need to be overcome before the potentially game-changing value of machine learning for petrophysics discipline can be realized. First, a robust theoretical foundation to support the application of machine leaning to petrophysical interpretation should be established; second, the utility of existing machine learning algorithms must be evaluated and tested in different petrophysical tasks with different data scenarios; third, procedures to control the quality of data used in machine leaning algorithms need to be implemented and the associated uncertainties need to be appropriately addressed. The paper will outlook the future opportunities of enabling advanced data analytics to solve challenging oilfield problems in the era of the 4th industrial revolution (IR4.0).

Rock typing in carbonate reservoirs is challenging due to high spatial heterogeneity and complex pore structure. In extreme cases, conventional rock typing methods such as Leverett's J-function, Winland's R 35 , and flow zone indicator are inadequate to capture the heterogeneity and complexity of carbonate petrofacies. Furthermore, these methods are based on core measurements, hence are not applicable to uncored reservoir zones. This paper introduces a new method for petrophysical rock classification in carbonate reservoirs that honors multiple well logs and emphasizes the signature of mud-filtrate invasion. The method classifies rocks based on both static and dynamic petrophysical properties. An inversion-based algorithm is implemented to simultaneously estimate mineralogy, porosity, and water saturation from well logs. We numerically simulate the process of mud-filtrate invasion in each rock type and quantify the corresponding effects on nuclear and resistivity measurements to derive invasion-induced well-log attributes, which are subsequently integrated into the rock classification. Under favorable conditions, the interpretation method advanced in this paper can distinguish bimodal from uni-modal behavior in saturation-dependent capillary pressure otherwise only possible with special core analysis. We successfully apply the new method to a mixed clastic-carbonate sequence in the Hugoton gas field, Kansas. Rock types derived with the new method are in good agreement with lithofacies described from core samples. The distribution of permeability and saturation estimated from well-log-derived rock types agrees with routine core measurements, with the corresponding uncertainty significantly reduced when compared to results obtained with conventional porosity-permeability correlations.

Summary Geoscientific and engineering experiments in petrophysics, rock physics, and rock mechanics depend on multiple, costly, and sometimes rare samples used to characterize the properties of natural rocks. Testing these samples helps in modeling various hydrocarbon recovery and stimulation scenarios, as well as understanding the fluid-rock interactions in the subsurface under various pressure and temperature conditions. Over the last decade, 3D printing has matured to become a more commonly available tool to enable repeatable experiments with controllable materials and pore system geometries to investigate petrophysical, geomechanical, and geophysical properties of porous rocks. This review introduces the development, characteristics, and capabilities of 3D printing technology that are specifically used in research. Applications in the realm of petrophysics highlight the issues of replicating the pore network geometry and subsurface physics, aiming at understanding fluid flow in porous media problems. Using 3D-printed models in rock mechanics experiments focuses on generating comparable geomechanical properties and reproducing fractures, joint surfaces, and other rock structures, whereas in rock physics, geophysical forward modeling is highlighted to take advantage of 3D printing technology. By summarizing the recent advances in 3D printing as applied to petrophysics, rock physics, and rock mechanics, this review paper presents the current state of the art and the challenges in scale, cost, time, and materials, as well as the directions for advancing this frontier discipline to answer various fundamental questions regarding porous media research using 3D printing technology.

No abstract

Petrophysical rock classification is an important component of the interpretation of core data and well logs acquired in complex reservoirs. Tight-gas sandstones exhibit large variability in all petrophysical properties due to complex pore topology resulting from diagenesis. Conventional methods that rely dominantly on hydraulic radius to classify and rank reservoir rocks are prone to rock misclassification at the low-porosity and lowpermeability end of the spectrum. We introduce a bimodal Gaussian density function to quantify complex pore systems in terms of pore volume, major pore-throat radii, and pore-throat radius uniformity. We define petrophysical dissimilarity (referred to as orthogonality) between two different pore systems by invoking the classic "bundle of capillary tubes" model and subsequently classify rocks by clustering an orthogonality matrix constructed with all available mercury injection capillary pressure data. The new method combines several rock textural attributes including porosity, pore-throat radius, and tortuosity for ranking reservoir rock quality in terms of flow capacity. We verify the new rock classification method with field data acquired in the Cotton Valley tight-gas sandstone reservoir located in the East Texas basin. The field case shows that the new method consistently identifies and ranks rock classes in various petrophysical data domains, including porositypermeability trends, pore-size distribution, mercury injection capillary pressure, and NMR transverse relaxation time (T 2) spectra. Relative permeability curves, which are difficult to measure in the laboratory for tight rocks, are quantified with Corey-Burdine's model using the bimodal Gaussian pore-size distribution and are validated with core data.

Compressional and shear sonic traveltime logs (DTC and DTS, respectively) are crucial for subsurface characterization and seismic-well tie. However, these two logs are often missing or incomplete in many oil and gas wells. Therefore, many petrophysical and geophysical workflows include sonic log synthetization or pseudo-log generation based on multivariate regression or rock physics relations. Started on March 1, 2020, and concluded on May 7, 2020, the SPWLA PDDA SIG hosted a contest aiming to predict the DTC and DTS logs from seven “easy-to-acquire” conventional logs using machine-learning methods (GitHub, 2020). In the contest, a total number of 20,525 data points with half-foot resolution from three wells was collected to train regression models using machine-learning techniques. Each data point had seven features, consisting of the conventional “easy-to-acquire” logs: caliper, neutron porosity, gamma ray (GR), deep resistivity, medium resistivity, photoelectric factor, and bulk density, respectively, as well as two sonic logs (DTC and DTS) as the target. The separate data set of 11,089 samples from a fourth well was then used as the blind test data set. The prediction performance of the model was evaluated using root mean square error (RMSE) as the metric, shown in the equation below: RMSE=sqrt(1/2*1/m* [∑_(i=1)^m▒〖(〖DTC〗_pred^i-〖DTC〗_true^i)〗^2 + 〖(〖DTS〗_pred^i-〖DTS〗_true^i)〗^2 ] In the benchmark model, (Yu et al., 2020), we used a Random Forest regressor and conducted minimal preprocessing to the training data set; an RMSE score of 17.93 was achieved on the test data set. The top five models from the contest, on average, beat the performance of our benchmark model by 27% in the RMSE score. In the paper, we will review these five solutions, including preprocess techniques and different machine-learning models, including neural network, long short-term memory (LSTM), and ensemble trees. We found that data cleaning and clustering were critical for improving the performance in all models.

We introduce a new non-parametric matched-filterbank spectral estimator, referred to as Amplitude and Phase Estimation (APES), to perform dispersion analysis of borehole array sonic measurements. This method extracts the dispersion characteristics of all wave modes by applying an APES filter to array sonic spectral data and converting the estimated wavenumber to slowness. The implemented adaptive filter in APES ensures that the output signal be sufficiently close to a sinusoid with a designated wavenumber in space domain, which constrains the interference from other wavenumber components and suppresses the noise gain. Consequently, the resolution and signal-noise-ratio of dispersion analysis is significantly enhanced. Dispersion fitness functions processed with APES indicate clearer and narrower ridges with minimum presence of alias. At each frequency, dispersions of all modes can be identified without knowledge a priori of the exact number of modes. More importantly, the new method is not computationally intensive compared to existing dispersion analysis methods. Processing examples with synthetic and field data are presented and compared with the weighted spectral semblance (WSS) method to demonstrate the applicability and advantages of this method.