Raster LogsMachine LearningWell Log Interpretation2023

VeerNet: Using Deep Neural Networks for Curve Classification and Digitization of Raster Well-Log Images

Publication Year

2023

BibTeX / Citation

@article{Nasim_2023, title={VeerNet: Using Deep Neural Networks for Curve Classification and Digitization of Raster Well-Log Images}, volume={9}, ISSN={2313-433X}, url={http://dx.doi.org/10.3390/jimaging9070136}, DOI={10.3390/jimaging9070136}, number={7}, journal={Journal of Imaging}, publisher={MDPI AG}, author={Nasim, M. Quamer and Patwardhan, Narendra and Maiti, Tannistha and Marrone, Stefano and Singh, Tarry}, year={2023}, month=July, pages={136} }

Visit Official Journal

Manuscript Preview

Open Fullscreen

Executive Summary

Raster logs are scanned representations of the analog data recorded in subsurface drilling. Geologists rely on these images to interpret well-log curves and deduce the physical properties of geological formations. Scanned images contain various artifacts, including hand-written texts, brightness variability, scan defects, etc. The manual effort involved in reading the data is substantial. To mitigate this, unsupervised computer vision techniques are employed to extract and interpret the curves digitally. Existing algorithms predominantly require manual intervention, resulting in slow processing times, and are erroneous. This research aims to address these challenges by proposing VeerNet, a deep neural network architecture designed to semantically segment the raster images from the background grid to classify and digitize (i.e., extracting the analytic formulation of the written curve) the well-log data. The proposed approach is based on a modified UNet-inspired architecture leveraging an attention-augmented read–process–write strategy to balance retaining key signals while dealing with the different input–output sizes. The reported results show that the proposed architecture efficiently classifies and digitizes the curves with an overall F1 score of 35% and Intersection over Union of 30%, achieving 97% recall and 0.11 Mean Absolute Error when compared with real data on binary segmentation of multiple curves. Finally, we analyzed VeerNet’s ability in predicting Gamma-ray values, achieving a Pearson coefficient score of 0.62 when compared to measured data.

Related Works

Efficient self-attention based joint optimization for lithology and petrophysical parameter estimation in the Athabasca Oil Sands

Accurately identifying lithology and petrophysical parameters, such as porosity and water saturation, are essential in reservoir characterization. Manual interpretation of well-log data, the conventional approach, is not only labor-intensive but also susceptible to human errors. To address these challenges of lithology identification and petrophysical parameter estimation in the Athabasca Oil Sands area, this study introduces an AutoRegressive Vision Transformer (ARViT) model for lithology and petrophysical parameter prediction. The effectiveness of ARViT lies in its self-attention mechanism and its ability to handle data sequentially, allowing the model to capture important spatial dependencies within the well-log data. This mechanism enables the model to identify subtle spatial and temporal relationships among various geophysical measurements. The model is also inter- pretable and can serve as an assistive tool for geoscientists, enabling faster interpretation while reducing human bias. The interpretable nature of the model should assist geoscientists in conducting faster quality checks of the predictions, ensuring that errors are not propagated to subsequent stages. This study adopts a multitask learning approach, jointly optimizing the model’s performance across multiple tasks simultaneously. To evaluate the effectiveness of the ARViT model, we conducted series of experiments and comparisions, testing it against traditional artificial neural networks (ANN), Long Short-Term Memory (LSTM), and Vision Transformer (ViT) models. To showcase the versatility of ARViT, we apply Low-Rank Adaptation (LoRA) to a different smaller dataset, showing its potential to adapt to different geological contexts. LoRA not only helps in model adaptability but also helps to reduce the number of trainable parameters. Our findings demonstrate that ARViT outperforms ANN, LSTM, and ViT in estimating lithological and petrophysical parameters. While lithology prediction has been a well-explored field, ARViT’s unique blend of features, including its self-attention mechanism, autore- gression, and multitask approach along with efficient fine tuning using LoRA, sets it apart as a valuable tool for the complex task of lithology prediction and petrophysical parameter estimation.

Automated Detection of Geological Features: Leveraging Deep Learning for Beddings and Fractures Identification in Image Logs

Fractures and bedding planes within geological formations influence the exploration and development of fields in the petroleum industry, playing a vital role in reservoir characterization. However, manually identifying these features from borehole image logs (BHIs) is labo- rious and time-consuming. This study addresses this challenge by proposing an automated system to streamline the fracture and bedding plane picking and identification process. The model introduced in this study aims to expedite result delivery while significantly reducing the manual workload by automating the detection of fractures and bedding planes. Trained on a data set of BHIs, including formation micro-imager (FMI) and compact micro-imager (CMI) from 14 wells in Oman, the method presented proposes a transformer-based mod- el, named Geological Beddings and Fractures Detection Transformer (GeoBFDT) for the identification, localization, and estimation of geological structural feature parameters. GeoBFDT uses a bipartite matching loss to match the predicted geological feature types and their depth, dip, and azimuth values with the ground truth in an end-to-end manner. Visual analyses, performance metrics, sensitivity tests, and comparative plots highlight the model's capability to identify subsurface features. Validation using various depth thresholds on the test set for fractures and beddings reveals F1-scores of ~65% (fractures) and ~63% (beddings) at a 3-cm depth threshold and ~75% (fractures) and ~69% (beddings) at a 5-cm threshold. Dip accuracy is ~79% (fractures) and ~65% (beddings) at a 1.5° dip threshold and ~90% for both fractures and beddings at a 3° dip threshold. Similarly, azimuth accuracy is at ~83% (fractures) and ~75% ( beddings) at a 7° azi- muth threshold and ~92% (fractures) and ~84% (beddings) for an azimuth threshold of 15°. The ablation study indicates that a smaller backbone yields better performance given the relatively small data set and that increasing the number of wells further enhances model performance. In addition, the dynamic logs of BHIs outperform static logs. The model's adaptability to different well inclinations was also assessed by evaluating its performance on fractures from five horizontal wells, achieving a ~55% score at a 4-cm threshold for fracture detection and localization tasks. Notably, the proposed approach surpasses the objectives of existing deep learning (DL)-based methods by accommodating multiple features in a single image, identifying beddings and fractures in complex geological areas, eliminating mask dependency, and providing an end-to-end strategy. Unlike models such as “you only look once” or mask region-based convolution neural network (R-CNN), which rely on mask generation, the proposed method focuses on different objectives, offering an alternative approach to feature picking. Although trained on limited data and validated through validation and testing sets, the GeoBFDT model offers an alter- native pathway toward an end-to-end automated geological feature picking tool from BHIs. We hope that in the future, this methodology can be further explored to automate the entire feature-picking workflow in an end-to-end fashion.