Depth matching well logs acquired from multiple logging passes in a single well has been a longstanding challenge for the industry. The existing approaches employed in commercial platforms are typically based on classical cross-correlation and covariance measures of two signals, followed by manual adjustments. These solutions do not satisfy the rising demand to minimize user intervention to proceed towards automated data interpretation. We aimed at developing a robust and fully automatic algorithm and workflow for depth matching gamma-ray logs, which are commonly used as a proxy to match the depth of other well logs measured in multiple logging passes within the same well. This was realized by a supervised machine learning approach through a fully connected neural network. The training dataset was obtained by manually labeling a limited set of field data. As it is unrealistic to expect a perfect model from the initial training with limited manually labeled data, we developed a continuously self-evolving depth-matching framework. During the use of depth-matching service, this framework allows taking the user input and feedback to further train and improve the depth-matching engines. This is facilitated by an automatic quality-control module for that we developed a dedicated metric by combining a few different algorithms. We use this metric to assess the quality of the returned results from the depth-matching engine. The users review the results and do manual adjustments if some intervals are not ideally depth matched by the engine. Those manual adjustments can be used to further improve the machine-learning model. A well-designed framework enables automatic and continuous self-evolving of the depth-matching service. A key aspect of the developed framework is its generalization potential because it is independent of the signal type. It could be easily extended for other log types, especially when the correlation thereof is not obvious, provided that a sufficiently large volume of labeled data is available. This framework has been prototyped and tested on field data.

Abstract Modern logging data are characterized by abundance and multiple dimensions: spatial, temporal, and physical. Traditional interpretation workflows for logging data are often limited in their scope and flexibility and use only a subset of the data dimensions, either by choice or necessity. Furthermore, the various deterministic or stochastic workflows typically rely on preconceived rock and fluid models. When those a priori models do not fit the data, confusion can ensue, whereas it is obvious that the correct model is buried somewhere in the “big logging data.” To overcome these limitations, new symbiotic approaches using domain expertise and data analytics (“domain-analytics”) have been developed. Domain experts use those novel approaches to develop data-driven workflows to explore and mine complex logging datasets for latent but interpretable information. Once validated to be idempotent, the expert workflows that respect both the acquired data and domain knowledge can be packaged into new classes of answers products. In this paper, the novel approaches, along with the examples of new answer products, are presented. These include 1) the automated spatial search for common modes and repeated patterns in the single or multi-well nuclear magnetic resonance (NMR) data, 2) the generation of a data-driven fluid model using time-lapse logging data acquired during multiple passes over the same formation, and 3) the creation of interpretive class-based models using definitive core data that could be propagated to continuous log data for a richer petrophysical interpretation.

Summary The paper illustrates the improvements in logging while drilling (LWD) images and subsequent formation evaluation by using a new methodology for depth and survey measurements corrections. LWD depth measurements are often considered inaccurate and, therefore, not as reliable for well-to-well correlations, correlations with data acquired with wireline measurements and formation layer thickness determinations. The reasons for these inaccuracies generally originate from the traditional practice that LWD depth is purposely made equal to the driller's depth, which is a static pipe length measurement made by tape at the surface. There is almost always a difference between the actual measured depth (MD) of the LWD sensor downhole and this static pipe measurement, because downhole the drillpipe is subject to an environment that is not representative of the derrick (e.g., varying drilling mechanical conditions and temperature changes). Here, we demonstrate the applications of the method, which allows dynamic driller's depth correction for the effects of drillstring weight, downhole friction, weight on bit, thermal expansion, residual rig heave, and tide. Another significant inaccuracy source is a standard practice of calculating borehole position from stationary survey points typically taken every 90 feet (ft) using the minimum curvature method. Neglecting the complex borehole shape between survey stations can lead to a systematic error in determining the borehole position. We consider using continuous inclination and azimuth measurements along with stationary surveys to correct these errors. We provide comparisons of LWD images before and after the depth and survey corrections to illustrate how the measurement errors affect formation dips interpreted from the images. We demonstrate how improved accuracy allows filtering out the artifacts and provides more decisive and accurate identification of geologic features. We show how using the corrected 3D position improves accuracy of the formation thicknesses calculations and therefore improves the reservoir summation results. As a result, we propose a borehole 3D position measurement that is accurate, consistent between wells (regardless of rig type or bottomhole assembly [BHA] configuration), and independent of the drilling mode. Using this new measurement significantly improves the quality of the formation evaluation.

An efficient method for determining regional dip angles and formation properties in high angle wells is presented. As a first step, logs are processed to detect similar features. The feature extraction process is speeded up by observing that variation in the formation properties of a high angle well is typically far less than that of a vertical well. Consequently, log data can be appropriately filtered and considerably down-sampled while preserving the essential features of the log. A wavelet-based method is used for multilevel decomposition of log data. Once similar features are recognized, tool trajectory and other information may be combined to select a few features that satisfy certain operating and geological constraints. These features, in conjunction with the logs and tool trajectory, provide an initial set of dip angles and formation parameters. Estimates of dip angles and formation parameters may be improved by an iterative procedure that minimizes the error between computed and measured logs.

This work describes a technique for segmentation of multiple logs to obtain common boundary. A principal component analysis is applied to further improve computation efficiency by exploiting inherent correlation among various measurements. A develop statistical approach for automatic termination of the algorithm is planned; that is, to have a priori estimation of number of final segments.

Selecting and designing the proper completion in naturally fractured reservoirs is always a challenge because of the mechanical and flow heterogeneities due to the fractures. Furthermore, when hydraulic fracturing is used to enhance the recovery, the interplay between the 3D stress field and the existing natural fracture systems becomes an important factor. Three mechanical scenarios might occur while fracturing the medium (Figure 1). First, the natural fractures may have no influence and the hydraulic fracture will propagate in a direction orthogonal to the minimum principal stress as expected in a classical model (Figure 1a). Second, only the system of natural fractures will be reactivated and eventually extended (Figure 1b). Third, both newly generated hydraulic fractures and natural fractures will intersect and propagate in a complex manner (Figure 1c). The tortuosity of hydraulic fractures will be greatly controlled by the anisotropy of the effective elastic medium due to 3D stress and natural fractures.