Exploratory analysis of sub-terahertz sensor data characteristics for the purposes of machine learning technique development

管道的绝缘下腐蚀（CUI）以及导致其产生的湿气是能源和加工行业面临的重大挑战。SubTera最近开发了一种改变游戏规则的管道检测能力，并在过去的18个月里，与全球能源生产商合作，对一种原型检测工具进行了现场测试。SubTera的技术可以通过多种方式帮助能源行业实现其净零排放目标。根据挪威石油安全局的数据，50%的陆上工厂报告的碳氢化合物泄漏是由CUI引起的。SubTera技术的主要功能是在最早的时候检测到CUI和湿度，这将降低管道故障的风险和成本。这有可能节省数百万英镑的新基础设施成本、清理费用和罚款；然而，相关的逸散性排放泄漏的减少将进一步减少我们的碳足迹。作为全球向净零能源世界过渡的一部分，再加上向工业4.0的发展，SubTera承认，在未来，其传感器技术必须集成到机器人平台中，并且检查将自动进行。在接下来的24个月里，SubTera计划开发一种新系统，该系统将结合机器学习并实现机器人集成。本文提出的探索性小型项目是实现这一未来的关键的第一步。在这个小型项目中，将分析使用SubTera的TRL7原型捕获的许多现有检测数据集（即：识别数据趋势、研究模式和变化、去除传感器噪声、理解错误和不确定性）。通过这一分析，将确定SubTera传感器的特性，并确定实现高效机器学习集成的要求。这个项目的输出将是一系列的见解，在一份报告中提出，以指导未来的系统设计（即：传感器，光学），系统操作，以及在SubTera的下一个开发阶段实现机器学习技术。

Corrosion Under Insulation (CUI) on pipework, and the moisture that causes it, are significant challenges faced by the energy and processing industries.SubTera has recently developed a game-changing pipework inspection capability, and over the past 18 months, has been field-testing a prototype inspection tool in collaboration with global energy producers.There are numerous ways in which SubTera's technology can contribute to helping the energy industry achieve its net-zero objectives. According to Norway's Petroleum Safety Authority, 50% of reported hydrocarbon leaks at onshore plants are caused by CUI. The primary function of SubTera's technology is to detect CUI and moisture at the earliest onset, which will reduce the risk of, and cost associated with, pipeline failure. This has the potential to save millions of pounds of new infrastructure costs, clean-up fees, and fines; however, the associated reduction in leakage of fugitive emissions will further reduce our carbon footprint.As part of the global transition to a net-zero energy world, coupled with the evolution toward Industry 4.0, SubTera acknowledges that in the future, its sensor technology must be integrated within robotic platforms, and inspections will be conducted autonomously. Over the next 24 months, SubTera plans to develop a new system, that incorporates machine learning and enables robotic integration. The exploratory mini-project proposed herein is a critical first step in enabling that future.During this mini-project, a number of existing inspection data sets, captured using SubTera's TRL7 prototype, will be analysed (i.e.: identifying data trends, study patterns and variation, removing sensor noise, understanding errors and uncertainties). Through this analysis, the characteristics of SubTera's sensors will be determined, and the requirements to enable efficient machine learning integration will be identified.The output from this project will be a series of insights, presented in a report, to guide future system design (i.e.: sensor, optical), system operation, and the implementation of machine learning techniques within SubTera's next development phase.