High Performance Processing of Distributed Acoustic Sensing Data

分布式声学传感数据的高性能处理

• 批准号: 2299114
• 金额: --
• 依托单位: University of Liverpool
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2019
• 资助国家: 英国
• 起止时间: 2019 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2299114
• 关键词: Performance; Processing; Distributed; Acoustic; Sensing

Distributed Acoustic Sensing (DAS) offers great potential to extract pertinent information from the environment close to optical fibres, turning each 50km long fibre into 5000 (or more!) acoustic microphones. This has significant commercial utility in contexts that include monitoring roads and railways as well as protecting critical infrastructure (eg gas pipelines and nuclear power stations). A market-leading DAS company have a novel ability to use coherent processing to maximise the ability to extract such information. This provides an opportunity to develop a novel Bayesian signal processing chain that fully exploits the novel sensing capability. This PhD will focus on developing a high performance variant of this processing chain using state-of-the-art techniques such as particle filters, Convolutional Neural Networks etc. The focus will be on a subset of: detection, localisation (e.g., using beamforming), tracking and classification of anomalies (e.g., the sounds of people walking, digging, driving, etc) as well as long term condition monitoring and simulation (both for assessment of performance and generation of training data for machine learning algorithms). Note that since the data-rates involved are high, it is anticipated that software will need to be developed with a view to implementation on a small cluster of GPUs or similar. The specific focus of the PhD will be chosen to be well matched to the skills of the student (as well as to the company's needs).

分布式声学传感(DAS)提供了从光纤附近的环境中提取相关信息的巨大潜力，将每条50公里长的光纤变成5000(或更多！)声学麦克风。这在监测公路和铁路以及保护关键基础设施(如天然气管道和核电站)等方面具有重要的商业效用。一家市场领先的DAS公司拥有一种新的能力，可以使用相干处理来最大限度地提高提取此类信息的能力。这为开发充分利用新型传感能力的新型贝叶斯信号处理链提供了机会。该博士学位将专注于使用最先进的技术，如粒子过滤器、卷积神经网络等，开发该处理链的高性能变体。重点将放在以下几个子集上：检测、定位(例如，使用波束成形)、异常(例如，人们行走、挖掘、驾驶的声音)的跟踪和分类，以及长期条件监测和模拟(用于评估性能和生成机器学习算法的训练数据)。注意，由于所涉及的数据率很高，预计将需要开发软件，以期在一小群图形处理器或类似设备上实施。博士学位的具体重点将与学生的技能(以及公司的需求)很好地匹配。