AI in Geophysical Surveys: Enhancing Data Acquisition, Processing, and Reporting

The geophysical industry, with its vast and intricate datasets, is ripe for the integration of Artificial Intelligence (AI). As the volume of geophysical data grows exponentially, thanks to advanced acquisition equipment, the challenge lies in efficiently processing, modeling, and predicting using this data. Traditional methods, while effective, often grapple with computational bottlenecks, especially with large-scale and complex geophysical systems. This is where AI, and more specifically Deep Learning (DL), steps in.

Stages of AI Implementation in Geophysical Surveys:

Data Acquisition:

Traditional Approach: Relies heavily on physical models and manual data collection.

?AI-Enhanced Approach: AI can handle vast datasets naturally, offering a computational advantage over traditional methods. Moreover, AI can utilize historical data and experience, which are typically overlooked in conventional methods. Advanced systems streamline the process of installing hull-mounted sensors, allowing for faster mobilization and high-speed surveys. Precise heading information enhances the accuracy of side scan sonar data, crucial for accurate target counting. Innovations in curved line surveying use advanced sensors to accurately survey even narrow corridors, reducing project schedules and emissions. Automated winch controls ensure the altitude of sensors is constantly controlled, improving data quality and paving the way for remote vessel operations.

Data Processing and Analysis:

Traditional Approach**: Often suffers from inaccurate modeling, requires intensive human labor, and faces computational challenges with large-scale systems.

?AI-Enhanced Approach: DL can provide high computational efficiency after training, enabling high-resolution characterization of the Earth. It doesn't necessitate an accurate description of the physical model, which is beneficial when the physical model is partially unknown or complex. For instance, in earthquake science, DL has improved earthquake detection accuracy to 100%, a significant leap from the 91% accuracy achieved with traditional methods.

Auto-Generated Survey Reports:

??- Traditional Approach: Time-consuming, requires expert knowledge, and often faces difficulties like geophysical inversion challenges.

??- AI-Enhanced Approach: DL can be applied to automate report generation, reducing the time and expertise required. By training on historical data, DL models can generate comprehensive and accurate reports, highlighting key insights and predictions.

Technical Advancements in AI for Data Acquisition:

The integration of AI into hydrographic operations has been a significant advancement. AI-driven tools provide automatic real-time data processing and navigation aiding. For instance, software tools trained on thousands of images can identify objects, making tasks like identifying coral species or monitoring mussel farms more efficient. Furthermore, software solutions are sensor-agnostic, integrating with virtually any sensor. They also provide real-time data display, data fusion, and automation of workflows and data processing. Solutions for unmanned surface vehicles (USVs), remotely operated vehicles (ROVs), and autonomous underwater vehicles (AUVs) are designed for specific survey challenges, complementing each other by being designed for surveys of different challenging environments and structures.

Deep Learning in Ground-Penetrating Radar:

Deep learning has achieved state-of-the-art performance on signal and image processing. Due to its remarkable success, it has been applied to more challenging tasks, such as ground-penetrating radar (GPR) testing in civil engineering. This technology has been employed to exploit signal and image data in GPR systems, achieving significant success. For instance, convolutional neural networks (CNNs) have been used for recognizing subgrade defects, characterizing and identifying asphalt mixtures, and even 3D reconstruction of concealed cracks. The combination of deep learning with GPR has revolutionized defect detection tasks, leading many groups to redesign their systems. The evolution of general processor units and the availability of large datasets have contributed to its recent tremendous success.

AI in Subsea Pipeline Inspection: Visual Data Analysis and Automated Anomaly Detection:

Subsea pipelines, crucial for the transportation of oil and gas products, are exposed to various environmental hazards. These hazards, such as mechanical damage and corrosion, can compromise the structural integrity of the pipelines, leading to potential catastrophic environmental and financial consequences. To ensure the safety and integrity of these pipelines, regular inspections are essential.

Automated Anomaly Detection:

Autonomous underwater systems (AUS) are increasingly being deployed to assist in subsea pipeline inspections. These systems are equipped with advanced sensors and cameras that capture real-time data. The challenge lies in analyzing this vast amount of data to detect potential pipeline damage. This is where computer vision techniques, combined with deep learning, come into play.

Computer Vision for Subsea Inspections:

Traditional inspection methods often rely on human operators to analyze the captured images and videos. This process can be time-consuming and prone to human errors.

With the integration of computer vision techniques, AUS can automatically analyze incoming images to detect anomalies. These techniques can identify pipeline features, damages like cracks in the coating, and even debris.

Object detection, a subset of computer vision, enables the identification of objects even in complex scenes where there might be occlusions or noise. This is particularly useful in subsea environments where visibility can be a challenge.

Deep Learning in Anomaly Detection:

Deep learning methods, especially those based on convolutional neural networks (CNNs), have shown state-of-the-art performance in various areas of computer vision, including image classification and object detection.

For subsea pipeline inspections, these methods can be trained to recognize specific features of a pipeline, such as anodes, or damages like cracks. They can also be trained to identify different types of debris and determine their nature.

One of the challenges in implementing deep learning for subsea inspections is the need for a large amount of labeled data. Labeling this data requires domain-specific expertise and can be both time-consuming and expensive. However, active learning techniques are being researched to optimize the labeling process, reducing costs while ensuring high performance of the object detectors.

Benefits of AI in Subsea Inspections:

Automated anomaly detection can significantly reduce the time required for inspections, leading to cost savings.

AI-driven inspections can achieve higher accuracy rates, reducing the chances of missing critical damages or anomalies.

The implementation of AI eliminates human errors that might occur during manual inspections, ensuring more reliable results.

The integration of AI in geophysical surveys offers a transformative approach to data acquisition, processing, and reporting. By harnessing the power of AI, geoscientists can achieve more accurate results, faster processing times, and comprehensive auto-generated reports, all while overcoming the limitations of traditional methods.

Reference:

https://www.sciencedirect.com/science/article/pii/S095006182032376X

Image-based and risk-informed detection of Subsea Pipeline damage https://link.springer.com/article/10.1007/s44163-023-00069-1

Digitalization for Subsea Asset Integrity Inspection: Bridging the gap between human operators and intelligent robots https://www.software.slb.com/blog/digitalization-for-subsea-asset-integrity-inspection