Anticipating geological formations prior to the advance of the drill bit is essential for optimizing drilling operations. This article introduces a novel design employing linear phased array acoustic transceivers to image subsurface structures ahead of the drill bit. The phased array, integrated into the drill collar as a transceiver device, scans the reservoir in a specific azimuth and utilizes echo signals to reconstruct the structure and properties of the underlying formation. To mitigate the impact of borehole model waves, we propose a semiblind source separation technique that isolates the predrilled boundary echo signal through residual error signal energy analysis. Numerical experiments demonstrate that the sector scan effectively visualizes the predrilled formation, with the reflected echo from geological boundaries enhanced by coherent accumulation using the receiving phased array. These results validate the feasibility of this method in identifying and alerting to predrilled formation boundaries.

Ultrasonic logging while drilling (ULWD) often faces challenges due to the complex downhole environment, instrument usage, and inevitable data compression, which significantly degrade the quality of logging images and introduce various noises. These factors impair the accuracy of geological analysis. To address this issue, we propose a novel multinoise ultrasonic logging image denoising diffusion method (MULDDM). This approach simplifies the training process for multiple types of logging noise by incorporating a logging multiple noise factor (LMNF), thereby significantly enhancing ULWD images quality. Additionally, to meet the deployment requirements of edge devices, we design a multistage progressive refinement network (MSPRN) to distill knowledge from MULDDM. This network reduces the model’s parameter count by 37.4% while maintaining excellent denoising performance during ULWD. Experimental results show that the MSPRN has a parameter size of just 22.7 M, with the signal-to-noise ratio of the denoised images exceeding 31 dB. The average processing time for a single logging image is approximately 0.1 s, supporting real-time image processing for logging edge equipment. This method effectively eliminates various types of logging noise while preserving crucial geological details, offering reliable data for accurate geological assessment.

Due to the limitations of the computing and transmission ability of the downhole edge devices, it is difficult to compress and transmit ultrasonic logging images, thus greatly degrading the efficiency of downhole imaging logging and monitoring. To achieve ultrasonic logging images compressed and reduce the cost of logging data transmission, a hardware-enabled compression sensing method is proposed. Firstly, an effective compressed sensing network is designed for downhole ultrasonic logging image compression. Then, the compressed sensing network is deployed on the edge device to compress the logging images. In addition, a reconstruction network is adopted to encode the compressed images and improve the reconstructed image quality. Finally, we evaluate the proposed method using real logging image datasets collected from the production environment. Experimental results demonstrate that the proposed method outperforms prior works in recovering logging images even at extremely low sampling ratios, meanwhile, it can effectively improve image reconstruction quality and reduce data transmission costs.

When conducting integrity assessments of metal well-bore casings in oil and gas wells using pulsed eddy current testing (PECT), the quantitative inversion of multi-layer wall thickness becomes challenging due to the complex effects of electromagnetic field propagation and eddy current coupling. We investigated the time-domain analytical solution of multi-layered sting for PECT and propose a time-window-integral (TWI) inversion method based on Tikhonov regular optimization. The correlation between numerical integration of time window and wall thickness reduction is analyzed according to eddy current response characteristics in time domain. Furthermore, the mathematical equation is established and corrected by regularization method. Finally, the simulation model of three-layer string is tested. The experimental results show that the average wall thickness error of each layer obtained by TWI inversion is 0.032mm, 0.049mm and 0.253mm respectively, and the average relative error is less than 5.5%, which can quickly obtain accurate wall thickness information.

To suppress noise and extract weak features in pulsed eddy current response signals, in this study, we employ the unscented Kalman filter (UKF) algorithm in conjunction with the radial basis function (RBF) neural network to establish the predictive and observational equations for the pulsed eddy current response signal. The RBF neural network's strong nonlinear capabilities are leveraged, while the UKF algorithm is integrated to mitigate noise and extract the signal. To assess the efficacy of this approach, we utilize COMSOL simulation data for validation purposes. Our findings demonstrate that the proposed algorithm effectively reduces noise, particularly for signals with a signal-to-noise ratio of 30 dB.

Given the limited computing, storage, and communication transmission abilities of downhole edge devices, it is awkward to transmit massive imaging data collected by an edge device to the ground server. The predicament results in long latency times and severely degrades the efficiency of downhole state monitoring. To reduce the cost of transmitting logging image data, a low-cost quantized compressed sensing (CS) and transmission framework is proposed in this work. The method exploits the inherent sparsity of the logging image data to compress and transmit the image as a 1-D vector. By significantly reducing the amount of data that needs to be transmitted, quantized CS and transmission (QCST) achieves efficient communication of information. Compared to existing compression reconstruction methods, a hardware-deployed implementation of a neural network (NN) is presented for image compression. Furthermore, a data quantization method is designed that convert the floating-point to binary data for reduced transmission costs. Finally, an image reconstruction network is adopted to improve the quality of compressed logging images using the powerful nonlinear modeling capabilities of NN. Comprehensive experiments demonstrate that the use of CS for compression and quantized transmission, combined with NN for reconstruction, could achieve a peak-signal-to-noise ratio (PSNR) of about 20 dB at extremely low sampling rates of 0.01 (1%) on a real-field dataset. It also outperforms existing methods in terms of maximizing compression gain and reconstruction quality, providing a new solution for logging data compression and transmission. Code available athttps://github.com/Qiang-Feng98/QCST.

Logging while drilling (LWD) is an efficient technique that provides real-time information about the reservoir while drilling. However, due to the limited transmission rate of the cables, only low-dimensional information can be transferred rather than high-dimensional logging images. This paper proposes an image compression method called the asymmetric convolutional autoencoder (ACAE) that can map raw logging data into low-dimensional embedding vectors for real-time transmission through the cables. The trained ACAE consists of two primary components: an encoder deployed onto the downhole computing module and a decoder deployed onto the ground computer. To decode low-dimensional embedded information with higher quality, attention modules are utilized in the decoder to suppress noise and attain smoother logging images. Experimental results demonstrate that our method can effectively compress original logging data with a small compression rate and reconstruct logging images.

The time-of-flight (ToF) and amplitude of an ultra-sonic echo signal are critical parameters employed to build a three-dimensional (3D) model of an underground borehole in an imaging logging system. However, due to different orientations of the initial measurement or repeated measurements, the point cloud generated by the ToFs of echo signals may not register the corresponding intensities produced by the signal amplitudes. This study proposes a borehole zone feature alignment framework to perform cross-modality registration between the point clouds of the borehole walls and their corresponding intensities. First, point clouds and intensities produced by the ultrasonic signal are mapped onto a plane, which is transformed into ToF and amplitude images to align the features of different modalities. The two-dimensional (2D)-to-3D registration is converted into a image matching problem. Then, we build a mathematical optimization model where the optimization target is to minimize the difference between ToF images and raw amplitude images. Finally, a multi-orientation search (MOS) method is proposed, which performs a search for the optimal shift matrices to achieve registration effectively and accurately. We manufactured an ultrasonic imaging logging tool and collected point cloud data for an entire production well to verify our proposed method. The experimental results demonstrated that the proposed method can carry out this cross-modality registration task effectively and yield intuitive models to help comprehend the construction of borehole walls in the well logging.

Time-of-flight (ToF) of ultrasonic waves is essential for petroleum well logging to draw borehole-wall images. This paper proposed a method that boosted accuracy of ToFs estimation in a complex geological environment. Unlike other classical methods, the proposed one adopts a one-dimensional convolutional neural network (1D-CNN) as a backbone to extract latent information related to ToF. For handling the shortage of ultrasonic waves with annotated ToFs, theoretical ultrasonic waves generated by manifold mathematical models are utilized as source domain to train the model, which is applied to estimate practical ultrasonic ToFs. Furthermore, since the distribution divergences exist between theoretical ultrasonic models and practical ones, Maximum Mean Discrepancy (MMD) and CORrelation ALignment (CORAL) as discrepancy loss functions are used to evaluate the distribution divergences between two domain datasets and to optimize the entire model. Tests on the ultrasonic waves acquired by an experimental well logging device demonstrate the proposed method has satisfactory performances.

The recognition and representation of natural fractures are essential since oil and gas can gather and flow through naturally fractured zones in the stratum. This paper aims to give an intuitive and precise fracture representation by three-dimensional (3D) models of borehole walls and proposes an unsupervised method to segment fracture zones in the 3D borehole wall models. First, the 3D point cloud of the borehole walls, obtained by an ultrasonic imaging logging system, is built to reveal the actual oil wells' structures intuitively. Subsequently, the fractured zones clustering (FZC) method is proposed to implement fracture recognition using the density-based spatial clustering, an unsupervised approach to deal with the shortage of annotated labels. Experiments on the well logging data of an actual production well demonstrate that the proposed method can identify the point cloud corresponding to fracture zones effectively and accurately.

Semantic recognition of fractures in well logging images is vital for engineers to implement oil and gas exploration. An essential approach to accomplishing the object is to adopt deep learning based on convolutional neural networks. However, due to the lack of annotated labels in well logging images, it is scarcely available to directly train the semantic segmentation network. In this paper, we explore a domain shift model attempting to achieve domain adaptation from one annotated dataset to our target well logging images. This model's core is to utilize adversarial learning, including generator and discriminator, which can prompt the model to generate fracture segmentation similar to the source domain in the target domain. For enhancing the domain adaptive model further, an attention module is introduced, which can suppress redundant noise in semantic segmentation results. We demonstrate that our proposed model performs well by extensive tests and ablation experiments in the ultrasonic well logging images.

Increasing the accuracy of the first arrival time extraction of the echo signal is the decisive technology of ultrasonic logging imaging. However, the performance of the existing first arrival time detection algorithms is inadequate, and wrong picking locations lead to indistinct local details and lack of scene adaptation, which makes it challenging to analyze and interpret the geological features. In this paper, a novel domain adaptation algorithm based on deep learning for one-dimensional time-domain ultrasonic signal reconstruction and first arrival time picking is proposed. We build the model called Deep Reconstruction-Arrival Picking Network (DRAPN), in which the source domain performs P-wave picking and waveform reconstruction on seismic signals, but the target domain detects the first arrival time of the unlabeled ultrasonic echo data. A formal analysis is provided to demonstrate the effectiveness of the algorithm in the domain adaptation context and imaging performance, which shows that DRAPN offers better accuracy overall than conventional algorithms.