Data-Augmented Deep Learning for Downhole Depth Sensing and Field Validation

TL;DR

This work tackles the challenge of accurate downhole depth calibration via casing collar locator signals in data-scarce environments. It presents the Signal Collection Vehicle (SCV) for downhole data acquisition and a comprehensive data-augmentation framework (normalization, LDS, LSR, geometric transformations, and multiple sampling) to train collar-recognition models. Two 1D CNN architectures, TAN and MAN, are evaluated as benchmarks for boundary-moment detection using probability maps rather than sparse one-hot labels. Systematic experiments reveal that standardization, LDS, and random cropping are fundamental, while LSR, time scaling, and multiple sampling significantly boost generalization, with real-well validation confirming accurate collar localization. The findings offer a practical pathway to improved downhole depth measurement in CCL-limited settings and highlight the value of tailored preprocessing in deep learning for geological signals.

Abstract

Accurate downhole depth measurement is essential for oil and gas well operations, directly influencing reservoir contact, production efficiency, and operational safety. Collar correlation using a casing collar locator (CCL) is fundamental for precise depth calibration. While neural network-based CCL signal recognition has achieved significant progress in collar identification, preprocessing methods for such applications remain underdeveloped. Moreover, the limited availability of real well data poses substantial challenges for training neural network models that require extensive datasets. This paper presents a system integrated into downhole tools for CCL signal acquisition to facilitate dataset construction. We propose comprehensive preprocessing methods for data augmentation and evaluate their effectiveness using our neural network models. Through systematic experimentation across various configuration combinations, we analyze the contribution of each augmentation method. Results demonstrate that standardization, label distribution smoothing (LDS), and random cropping are fundamental requirements for model training, while label smoothing regularization (LSR), time scaling, and multiple sampling significantly enhance model generalization capability. The F1 scores of our two benchmark models trained with the proposed augmentation methods maximumly improve from 0.937 and 0.952 to 1.0 and 1.0, respectively. Performance validation on real CCL waveforms confirms the effectiveness and practical applicability of our approach. This work addresses the gaps in data augmentation methodologies for training casing collar recognition models in CCL data-limited environments.

Figures (5)

• Figure 1: Cross-sectional illustration of a typical oil and gas well structure. Representative casing collar signatures from magnetic response are illustrated in dark blue near the corresponding casing collar, while the typical interference signals distant from casing collars are illustrated in dark green. Refer to figures from xiao2025realization.
• Figure 2: Structure of downhole tools integrated with the Signal Collection Vehicle (SCV): (a) Schematic diagram of the internal structure of the perforating gun employed in this work; (b) Functional structure diagram of the SCV; (c) A SCV circuit board.
• Figure 3: The train and inference process of this work: (a) Segment normalized CCL waveform with manually labeled collar marks, each segment containing a collar mark at its center (signal shapes in boxes are illustrative only); (b) Augmentation methods for preprocessing waveform segments; (c)-(d) Neural network architectures employed in this work; (e) Multiple random augmentations applied to each segment generate diverse sub-segment variants for training and testing datasets; (f) The procedure of casing collar recognition from CCL waveforms using sliding windows with overlap.
• Figure 4: Evaluation metrics of training progress under different configurations, including cross-entropy loss, F1 score, and area under the precision-recall curve (AUC-PR). The curve indexes correspond to configurations in Tables \ref{['tab1']} and \ref{['tab2']}. The meanings of abbreviations refer to Tables \ref{['tab1']} and \ref{['tab2']}. (a)-(c) Evaluation metrics for Group 1 configurations; (d)-(f) Evaluation metrics for Group 2 configurations; (g)-(i) Enlarged sections of (d)-(f); (j)-(l) Evaluation metrics for Group 3 configurations; (m)-(o) Enlarged sections of (j)-(l); (p)-(r) legends for (a)-(c), (d)-(i), and (j)-(o), respectively.
• Figure 5: Probability maps and recognition results for various configurations. (a) Cfg. 1 (using OHE); (b) Cfg. 2 (using fixed cropping); (c) Cfg. 3 (using LDS and random cropping); (d) Cfg. 6 (using fundamental methods with MAN model); (e) Cfg. 13 (optimal combination candidate using TAN model); (f) Full results of Cfg. 14 (optimal combination candidate using MAN model).