Multi-Modal ILI Data Fusion for Combined Diagnostics of Pipeline

金属损耗和隐蔽缺陷是油气产品输送管道的重要威胁。管道在线检测是利用无损检测技术来评价和维护管道在整个生命周期内的结构完整性。然而，由于单个无损检测技术的局限性，通常会应用多种无损检测方法，综合解释组合检测结果将实现精确的完整性分析。罗森的漏磁（MFL）和超声壁测量（UTWM）检测工具旨在将高水平磁化与高功率超声波结合应用于管道。这两种无损检测应用的组合结合了两种技术中的最佳技术，以确保管道在整个使用寿命期间的结构完整性。然而，解释和分析ILI数据的自动化过程对于准确和可靠的状况评估至关重要。本研究计画旨在发展分析演算法，以融合漏磁与超声波检测资料，进行管道完整性问题之联合诊断。本研究将首先集中在ILI数据的不确定性分析。然后，我们将使用深度学习方法将ILI数据从一个域转换到另一个域，例如，将MFL数据转换为UTMW数据，反之亦然。利用不确定性的知识，可以基于显著特征来对齐或配准翻译的ILI数据，并与自动编码器网络融合。研究成果将被集成到罗森的RoCombo MFL/UTWM检测系统中，以改善相互作用缺陷的表征。因此，综合检测与数据分析可以获得可靠的结果，以支持管道管理的决策。

Metal loss and hidden flaws are critical threats to the pipeline for transporting oil and gas products. Non-destructive testing (NDT) techniques are employed as the in-line inspection (ILI) to evaluate and maintain the structural integrity during the pipeline's entire lifecycle. However, due to the limitation of the individual NDT technique, multiple NDT methods are often applied, and a comprehensive interpretation of the inspection results from the combination will achieve a precise integrity analysis. ROSEN's magnetic flux leakage (MFL) and ultrasonic wall measurement (UTWM) inspection tools are designed to apply high-level magnetization in combination with high-power ultrasonic waves to the pipeline. This combination of two NDT applications combines the best of the two techniques to ensure the pipeline's structural integrity during its entire lifespan. However, an automated process to interpret and analyze the ILI data is critical for accurate and reliable condition assessment. This research project is to develop analytic algorithms to fuse the MFL and UTWM data for the combined diagnosis of pipeline integrity issues. The research will first focus on the uncertainty analysis of the ILI data. Then, we will convert ILI data from one domain to another with deep learning approaches, for example, translating MFL data to UTMW data and vice versa. With the knowledge of the uncertainty, the translated ILI data can be aligned or registered based on the salient features and fused with the autoencoder network. The outcomes of the research will be integrated into ROSEN's RoCombo MFL/UTWM inspection system for improved characterization of interacting defects. Thus, the comprehensive inspection with data analytics can achieve reliable results to support the decision-making for pipeline management.