METHOD FOR EVALUATING THE QUALITY OF LWD IMAGE PROFILES

Description

1 / 18 METHOD FOR EVALUATING THE QUALITY OF LWD IMAGE PROFILES Field of invention

[001] The present invention is situated in the technical field of reservoir modeling, simulation and evaluation. More specifically, the present invention relates to a method for evaluating the quality of LWD (Logging While Drilling) image profiles. Fundamentals of the invention

[002] The acquisition of petrophysical data in oil wells is a primary activity to support the proper development of an oil field.

[003] By acquiring data on the rocks drilled during well drilling, as well as data on the fluids contained in these rocks, it becomes possible to develop a field development strategy aimed at optimizing hydrocarbon recovery over time. Additionally, the information contained in the petrophysical data is of great value in making decisions regarding well completion.

[004] In this sense, the acquisition of petrophysical data can be carried out in two ways: the first, called LWD (Logging While Drilling), is performed during well drilling, using tools positioned in the drill string, close to the drill bit; the second, called wireline logging, uses wireline tools that are lowered into the well after drilling is completed and before the production casing is installed.

[005] In particular, after the acquisition of petrophysical data is completed, this data is sent to specialists in Petition 870220121268, dated 12 / 22 / 2022, page 19 / 45 2 / 18 subject, which involves processing and interpreting the data, subsequently incorporating the information into geological models of the fields.

[006] The technological advancements achieved by logging tools have generally allowed petrophysical data to be acquired using both LWD tools and wireline tools. Therefore, as a rule, it is possible for one type of operation to completely replace the other.

[007] In particular, in scenarios where the data of interest can be obtained with quality during the well drilling itself, using LWD tools, there are significant financial and time gains for a company. This is because LWD acquisitions have three important characteristics that distinguish them from wireline acquisitions: information about the drilled rocks is known in real time, which allows operational decisions about the well trajectory and length to be made more quickly and with a higher degree of reliability; the drill string allows LWD logging tools to reach high-inclination or even horizontal sections of the well with relative ease, which is not true for wireline-down tools; and string acquisition does not consume rig time in addition to the drilling time itself, as the activities are carried out concurrently.

[008] In contrast, data acquisition via wireline logging requires the assembly and disassembly of the logging rig in the rotary table area, time for lowering and passing the wireline logging tools in front of the Petition 870220121268, dated 12 / 22 / 2022, page 20 / 45 3 / 18 reservoir, as well as additional time for well conditioning after logging is completed. Therefore, it is possible to note that, in addition to delaying the completion of well construction and, consequently, the start of the next planned rig intervention, the additional time also directly impacts the operation's budget, as it assigns to the well the cost of rig rental during this period, for example.

[009] However, certain petrophysical data, although available through acquisition via both LWD logging and wireline logging, tend to have better quality when acquired via wireline logging, as the logging environment for this operation is much less noisy than that encountered during well drilling, where LWD logging is used. In addition, during wireline logging, the well presents better conditions for data acquisition, therefore having fewer suspended solids and a more homogeneous drilling fluid.

[010] Considering the fact that petrophysical data acquisition by wireline logging occurs after acquisition by LWD logging, in principle, it would be possible to opt for additional wireline logging to acquire data again containing a resistive image profile when its quality is not satisfactory when acquired by LWD logging. In this way, it would be possible to guarantee the acquisition of good quality data in the well under construction.

[011] However, in practice, it is not feasible to manually assess the quality of the aforementioned image profile acquired with LWD profiling in a timely manner to Petition 870220121268, dated 12 / 22 / 2022, page 21 / 45 4 / 18 decide on the need to make an additional purchase through cable profiling.

[012] In particular, an image profile has a large amount of data and requires a long time to be processed and analyzed by a specialist in the technique, a time that is longer than that normally available between the two data acquisition operations.

[013] An alternative solution, consisting of obtaining the image profile via cable profiling in a conservative manner, without prior evaluation of the quality of the image profile obtained by LWD profiling, could lead to the waste of a company's financial and physical resources.

[014] Based on these limitations, there is a need to automate the process of evaluating the quality of resistive image profiles acquired via LWD profiling, in order to make it feasible to perform this process in time to decide whether it is necessary to acquire the same resistive image profile via wireline profiling.

[015] Furthermore, there is a clear need for a solution that can analyze criteria subject to the subjectivity of human interpretation with regard to the quality of resistive image profiles. State of the art

[016] In the state of the art, there are some solutions involving the use of machine learning and artificial intelligence techniques applied to the resistive image profile acquired by LWD profiling, as presented in the following document.

[017] Document US2022090481A1 describes a method and Petition 870220121268, dated 12 / 22 / 2022, page 22 / 45 5 / 18 system for processing LWD image profile data, generating image data relative to the region of interest using a neural network, while well profile data is being acquired in the well. Furthermore, this document describes the transmission, based on the drilling parameter, of a command to a drilling system coupled to the well. However, this solution does not cover the evaluation of the quality of resistive image profiles acquired via LWD logging.

[018] Thus, there is a need for a method to automatically evaluate the quality of LWD image profiles, allowing for timely and efficient decision-making regarding the need to use cable profiling.

[019] Furthermore, there is a clear need for a solution that can analyze criteria subject to the subjectivity of human interpretation with regard to the quality of resistive image profiles. Brief description of the invention

[020] The present invention relates to a method for evaluating the quality of LWD (Logging While Drilling) image profiles comprising processing a plurality of LWD image profiles; subdividing LWD image profiles into smaller pseudo-images of the same size; normalizing each pseudo-image of the LWD image profiles; and classifying the LWD image profile according to its quality, which comprises classifying a plurality of segments of the LWD image profile into three quality categories, including: good, medium or poor, using a neural network model trained for evaluation. Petition 870220121268, dated 12 / 22 / 2022, p. 23 / 45 6 / 18 of the LWD image profile quality.

[021] Furthermore, according to a preferred embodiment of the present invention, the step of performing the processing of a plurality of LWD image profiles comprises removing spurious points from the LWD image profiles.

[022] Additionally, the step of performing the processing of a plurality of LWD image profiles further involves defining a single classification window of 120 rows and 120 columns.

[023] The method for evaluating the quality of LWD image profiles preferably also includes performing data augmentation.

[024] In accordance with a preferred embodiment, the good quality category comprises a section where it is possible to identify boundaries between rock layers, as well as textural elements internal to the layers; the medium quality category comprises a section where it is possible to identify boundaries between rock layers, but it is not possible to identify internal elements within them; and the poor quality category comprises a section lacking boundaries between layers or their internal characteristics.

[025] Furthermore, according to a preferred embodiment, the neural network model for evaluating the quality of the LWD image profile comprises a convolutional neural network architecture followed by a direct network; more preferably, the convolutional neural network is a VGG16 network.

[026] Based on the limitations of the prior art, the present invention automates the method for evaluating the quality of LWD image profiles with the aim of enabling Petition 870220121268, dated 12 / 22 / 2022, page 24 / 45 7 / 18 its completion in time to decide on the need to acquire the image profile via cable profiling tool. Brief description of the figures

[027] In order to complement the present description and obtain a better understanding of the characteristics of the present invention, and in accordance with a preferred embodiment thereof, a set of figures is presented, where, in an exemplary, though not limiting, manner, its preferred embodiment is represented.

[028] Figure 1 illustrates the result of the LWD image profile quality classification for a well.

[029] Figure 2 represents the second processing stage of a plurality of LWD image profiles.

[030] Figure 3 shows the data augmentation procedure.

[031] Figure 4 illustrates a training graph of the neural network model.

[032] Figure 5.1 shows the result of a blind test, comparing the response of the neural network model with the responses on the classifications given by expert petrophysicists.

[033] Figure 5.2 highlights the evaluation metrics of the neural network model trained to perform the method of the present invention. Detailed description of the invention

[034] The present invention relates to a method for evaluating the quality of LWD (Logging While Drilling) image profiles, comprising the steps as described below. Petition 870220121268, dated 12 / 22 / 2022, page 25 / 45 8 / 18

[035] According to a preferred embodiment of the present invention, the first step of analyzing a plurality of LWD image profiles acquired using an LWD (Logging While Drilling) logging tool can be performed by one or more specialists. In this step, preferably, at least 5,000 meters of LWD image profile are used, adding the data from all existing wells in the dataset.

[036] Specifically, the first step in analyzing a plurality of image profiles involves classifying a plurality of segments from the LWD image profiles into one of three quality categories, namely: - good - section where it is possible to identify boundaries between rock layers, as well as textural elements internal to the layers; - medium - section where it is possible to identify boundaries between rock layers, but it is not possible to identify internal elements within them; and - bad - section lacking boundaries between layers or their internal characteristics.

[037] In this sense, there are no requirements regarding the length of the classified sections, that is, in a single profile there may be short sections classified in the good category and longer sections classified in the medium category, for example.

[038] Thus, the result of the first step described above is the classified LWD image profiles.

[039] For illustrative and exemplary purposes, figure 1 shows a visualization of the result of the first stage of analyzing a plurality of LWD image profiles, where each quality category was associated with a color. Petition 870220121268, dated 12 / 22 / 2022, page 26 / 45 9 / 18

[040] The image profiles provide 360° coverage of the wellbore wall with a vertical resolution of 0.5 cm. The LWD tool operates by emitting acoustic energy and then measuring the amplitude and transit time of the reflected signal. For this, there are numerous sensors around the LWD tool (around 100 to 120 channels) to perform the wellbore wall reading.

[041] The data resulting from acquisition by the LWD tool is a distribution matrix with a number of columns equivalent to the number of channels existing in the LWD tool and the number of rows equal to the profiled length divided by the vertical resolution of the tool.

[042] The second stage involves processing a plurality of LWD image profiles, as well as the plurality of LWD image profiles classified in the first stage, which include a plurality of segments classified into three quality categories: good, medium or poor.

[043] More specifically, the second step in performing the processing of a plurality of LWD image profiles involves removing spurious points from the LWD image profiles and / or from the LWD image profiles classified in the three quality categories.

[044] Due to the fact that reservoirs have different thicknesses, the data acquisition rate in different profiling operations may vary, and if one wishes to profile only specific sections of a given reservoir, each image profile may have a different extent. That is, although the number of data columns remains constant, the number of rows will depend on the Petition 870220121268, dated 12 / 22 / 2022, page 27 / 45 10 / 18 conditions of each acquisition. On the other hand, the dimensions for data input in a neural network are of fixed size. Furthermore, the number of lines in an image profile can reach hundreds of thousands, which means an order of magnitude greater than the conventional size of neural networks, and larger sizes imply greater computational effort for training and using neural networks.

[045] Therefore, the processing step for a plurality of LWD image profiles additionally involves defining a single classification window of 120 rows and 120 columns. In this way, it involves subdividing the LWD image profiles and / or the classified LWD image profiles into smaller pseudo-images of the same size.

[046] Next, the second step involves defining a single quality classification, which is predominantly found in that section, for the pseudoimages of the classified LWD image profiles. Attention must be paid to the overall balance of the data in the database, that is, the number of sections with good, average and poor classifications should be similar, as illustrated in Figure 2.

[047] The third step of performing the normalization of each pseudo-image of the LWD image profiles and / or the classified LWD image profiles, according to the distribution of values ​​found in their own value matrix, results in the absence of disparities in the distribution range in the images used in the fourth step, namely performing the training of the LWD image profile quality assessment model. Petition 870220121268, dated 12 / 22 / 2022, page 28 / 45 11 / 18

[048] Furthermore, the third step in normalizing each pseudo-image of the LWD image profiles and / or the classified LWD image profiles involves performing data augmentation in order to increase the amount of training data, based on the data that already exists in the database. In particular, the data augmentation procedure of mirroring all the images generated so far is used, that is, the LWD image profiles and / or the classified LWD image profiles, as illustrated in Figure 3.

[049] The fourth step involves training a neural network model to evaluate the quality of the LWD image profile.

[050] In this sense, for a neural network to be able to correctly interpret the content of the LWD image profiles to be classified, it first needs to be trained for this purpose. In the context of neural networks, this consists of providing a series of images as input to the network and comparing the classification returned by it with the already known real classifications of these same images. Based on the errors committed by the neural network, its internal parameters (weights associated with the neurons) are adjusted by means of an optimization algorithm, which seeks to minimize these errors in order to improve the responses until a satisfactory level of accuracy is reached.

[051] Specifically, the fourth step in training the LWD image profile quality assessment model involves training the neural network model for LWD image profile quality assessment from Petition 870220121268, dated 12 / 22 / 2022, page 29 / 45 12 / 18 entries including LWD image profiles and classified LWD image profiles, which went through the second processing stage of a plurality of LWD image profiles and the third stage of performing normalization of each pseudo-image of the LWD image profiles and / or the classified LWD image profiles.

[052] Furthermore, the neural network model for evaluating the quality of the LWD image profile comprises a convolutional neural network architecture, specifically, a VGG16 network at its beginning and then the output of this network feeds into a direct network (FCN, Fully Convolutional Network), where the output of this latter network is compared with the corresponding classification.

[053] Regarding the VGG16 convolutional neural network, the first two layers have 64 channels of filter size 3 * 3 and the same padding. Then, after a pass-max pool layer (2, 2), there are two layers that have convolution layers of filter size 256 and filter size (3, 3); followed by a pass-max accumulation layer (2, 2), which is the same as the previous layer. In this way, they comprise 2 convolution layers of filter size (3, 3) and filter size 256. After that, there are 2 sets of 3 convolution layers and one pass-max pool layer. Each has 512 filters of size (3, 3) with the same padding. This image is then passed to the stack of two convolution layers.

[054] With regard to the direct network, this includes two layers, where the first layer contains 2048 nodes and the second contains 1024 nodes and where each node is followed by the Rectified Linear Units function for its activation. Petition 870220121268, dated 12 / 22 / 2022, pp. 30 / 45 13 / 18

[055] After the fourth stage of training a neural network model for evaluating the quality of the LWD image profile, the neural network model trained for evaluating the quality of the LWD image profile is able to classify an LWD image profile according to its quality, specifically in the three quality categories, good, medium or poor.

[056] The fifth step of classifying the LWD image profile according to its quality involves using the neural network model trained to evaluate the quality of the LWD image profile, which is the result of the fourth step. Specifically, the step of classifying the LWD image profile according to its quality involves classifying a plurality of sections of the LWD image profile into three quality categories, including: good, medium, or poor.

[057] Specifically, the three quality categories are: - good - section where it is possible to identify boundaries between rock layers, as well as textural elements internal to the layers; - medium - section where it is possible to identify boundaries between rock layers, but it is not possible to identify internal elements within them; and - bad - section lacking boundaries between layers or their internal characteristics.

[058] In particular, the fifth step of classifying the LWD image profile according to its quality occurs after the second step of processing a plurality of LWD image profiles and the third step of performing the normalization of each pseudo-image of the LWD image profiles. Petition 870220121268, dated 12 / 22 / 2022, page 31 / 45 14 / 18 and / or classified LWD image profiles.

[059] Specifically, the fifth step of classifying the LWD image profile according to its quality involves using a trained neural network model to evaluate the quality of the LWD image profile.

[060] In particular, the method for evaluating the quality of LWD image profiles can be used directly in drilling location in order to immediately obtain the automated classification of the LWD image profile being acquired and thus assist in decision-making. Illustratively and exemplarily, immediately after the LWD logging operation, still during well drilling, the collected data are received on the surface at the rig and transmitted to professionals in the onshore office to perform their quality classification. For this purpose, the data can be loaded into commercial software, such as petrophysics software, where the code or set of instructions comprising the trained neural network model that performs the method for evaluating the quality of LWD image profiles of the present invention is then run, which generates a quality profile of that loaded image in an expedited and direct manner.Based on this quality profile and the intended objectives for the well, the petrophysicists responsible for monitoring the acquisition can decide on the best operational sequence to adopt: perform a new LWD image logging, perform a wireline image logging, or not acquire additional image data.

[061] As a consequence of the ability to assess the quality of the LWD image profile in real time, in a way Petition 870220121268, dated 12 / 22 / 2022, page 32 / 45 With automatic 15 / 18 systems, unnecessary expenditure of financial resources on redundant acquisitions is avoided, as well as the unnecessary provision of wired imaging tools, which can become a critical resource. Additionally, it also prevents the well data acquisition phase from being completed without acquiring a satisfactory quality image, necessary for the proper development of the production project and for optimized well completion.

[062] In an example of an embodiment of the method according to the present invention, starting from the second stage of processing the data generated by the experts, the image profiles of 11 different wells were subdivided into 6868 pseudo-images to perform the fourth stage of training the LWD image profile quality assessment model, generating fixed-dimension inputs for the neural network and a significant number of individual images for training (an important factor for network performance), as well as maintaining a reasonable amount of information in each image so that the network has the ability to make distinctions between them.

[063] Additionally, by way of example, according to a preferred embodiment of the present invention, variations of neural networks and parameters were tested, as presented in Table 1 below. Table 1 - Example of parameters and structures of neural networks tested in different combinations. CNN FCN Activation LR Optimizer ResNet 50 1024 Sigmoid id 2.00E-2 Adam VGG16 2048-1024 Sigmoid Sigmoid id 2.00E-3 SGD Inception 1024-512 ReLu-ReLu 2.00E-4 Xception 2.00E-5 Petition 870220121268, dated 12 / 22 / 2022, pp. 33 / 45 16 / 18

[064] In Table 1, the CNN column represents the type of convolutional network used, the FCN column represents the number and size of the layers of the forward neural network coupled to the convolutional network, the activation column represents the activation functions used in each layer of the forward network, the LR column represents the learning rate used during training, and the optimizer column represents the search method used in the training process.

[065] Regarding performance, by way of example, to speed up the training process, a set of predefined weights, available in the tensorflow library, was used to initialize the convolutional network, where the weights are obtained from trainings performed using the free ImageNet repository as a database. Additionally, a separate pre-training of the forward neural network (FCN) was performed, with the aim of accelerating the subsequent global training of the LWD image profile quality assessment model. As a result, according to an example, an average of 16 epochs were required for global training, considering all tested configurations. Furthermore, in order to avoid overfitting problems with the available data, the model selected for each configuration was the one corresponding to the smallest error calculated based on the validation data.

[066] Figure 4 presents a training graph of the neural network model to perform the method for evaluating the quality of LWD image profiles, specifically the neural network model configuration that showed the best performance among all combinations. Petition 870220121268, dated 12 / 22 / 2022, pp. 34 / 45 17 / 18 tested. As can be seen in Figure 4, the best performing configuration stood out for a VGG16 convolutional neural network connected to a 2-layer forward network with sizes 2048 and 1024 and ReLU activation functions, using a learning rate of 2E-5 and the Adam optimizer.

[067] Other mechanisms, such as subdividing the dataset into batches and varying the learning rate during training itself, can also be used to optimize the process (i.e., speed up training and obtain better results in terms of error reduction).

[068] To demonstrate, by way of example, the accuracy and robustness of the model trained to perform the method for evaluating the quality of LWD image profiles, a visualization of the blind test is presented in Figure 5.1, comparing the response of said model with the responses on the classifications given by expert petrophysicists, which were used to train said model. In addition, in Figure 5.2 and Table 2, a summary of the main metrics for evaluating machine learning models is presented: precision, recall, accuracy, ROC and AUC, which were used to evaluate the model trained to perform the method of the present invention. Table 2 - Results of the blind test performed to evaluate the model training. Class Precision Recall Accuracy Samples 1 92.7% 88.4% 90.5% 43 10 75.6% 77.3% 76.4% 44 100 84.6% 86.3% 85.4% 51 Macro Average 84.2% 84.0% 84.1% 138 Weighted Average 84.2% 84.1% 84.1% 138 Petition 870220121268, dated 12 / 22 / 2022, pages 35 / 45 18 / 18

[069] Those skilled in the art in the technical field of mechanical engineering will appreciate the knowledge presented here and will be able to reproduce the invention in the embodiments presented and in other variants, covered within the scope of the appended claims. Petition 870220121268, dated 12 / 22 / 2022, pages 36 / 45

Claims

1 / 2 CLAIMS 1. A method for evaluating the quality of LWD (Logging While Drilling) image profiles, characterized by comprising: processing a plurality of LWD image profiles; subdividing the LWD image profiles into smaller pseudo-images of the same size; normalizing each pseudo-image of the LWD image profiles; and classifying the LWD image profile according to its quality, which comprises classifying a plurality of segments of the LWD image profile into three quality categories, including: good, medium, or poor, using a neural network model trained for evaluating the quality of the LWD image profile.

2. A method according to claim 1, characterized in that the step of processing a plurality of LWD image profiles comprises removing spurious points from the LWD image profiles.

3. Method, according to claim 1 or 2, characterized in that the step of performing the processing of a plurality of LWD image profiles additionally comprises defining a single classification window of 120 rows and 120 columns.

4. Method according to claim 1, characterized in that it further comprises performing data augmentation.

5. Method, according to claim 1, characterized in that the good quality category Petition 870220121268, of 12 / 22 / 2022, page 37 / 45 2 / 2 comprises a section where it is possible to identify boundaries between rock layers, as well as textural elements internal to the layers; the medium quality category comprises a section where it is possible to identify boundaries between rock layers, but it is not possible to identify internal elements within them; and the poor quality category comprises a section lacking boundaries between layers or their internal characteristics.

6. Method, according to claim 1, characterized in that the neural network model for evaluating the quality of the LWD image profile comprises a convolutional neural network architecture followed by a direct network.

7. Method, according to claim 6, characterized in that the convolutional neural network is a VGG16 network. Petition 870220121268, dated 12 / 22 / 2022, pp. 38 / 45

Images