BDD: A Big-Data Computational Laboratory for the Optimization of Olfactory Search Algorithms in Turbulent Environments

自然灾害、污染源和恶意恐怖主义行为的有害影响蔓延到影响点之外，威胁到更大区域的安全与保障。流体湍流是这些事件的同谋：它迅速将污染物或从源头释放的物质分散到大气或海洋中，其漩涡和涡流将分散的物质带离源头。当源的位置是未知的，湍流也混淆了远程传感器推断源位置的能力。该项目利用大数据策略开发新的搜索算法，可以使用固定或移动的传感器识别在湍流空气或海洋中释放的污染物或其他物质的来源。主要目标是直接提高我们的准备，以应对不良事件，当释放的药剂威胁到人类或环境安全。识别复杂湍流中污染物来源和扩散模式的能力具有重要的环境和安全意义。在不知道震源强度的情况下，必须从远程测量中推断。当污染源的位置和强度都未知时，跟踪标量源是一个不适定问题，因此是最具挑战性的。在应对环境和安全威胁方面，这些任务往往也是最紧迫的。在这种情况下的准备需要事先规划和自适应实时能力。因此，为了准确有效地识别污染物的位置和强度，设计了固定传感器的最佳位置和嗅觉搜索算法——这是一个艰巨的工程挑战，因为探测器上的信号高度间歇，前驱湍流弥散路径的复杂性。解码这些信息的最佳策略是使用大数据计算实验室制定的，在该实验室中，各种湍流的整个时空演化，通过高保真直接数值模拟计算得出，可以实时探测。源识别算法是一个利用伴随场的性质来识别源位置的变分问题。变分方法提供了一个自然的框架来优化固定传感器在感兴趣区域的位置，并引导嗅觉机器人向源运动。利用可公开访问的计算实验室、不同的流场配置和不同级别的流场保真度进行的重复虚拟实验，提供了对存在尺度依赖模型不确定性的搜索策略性能的准确评估。使用自主水下航行器对搜索策略的实际性能进行了测试。

The detrimental effects of natural disasters, sources of pollution and malicious acts of terrorism spread beyond the point of impact and threaten safety and security in a much larger region. Fluid turbulence is complicit in these events: It quickly disperses pollutants or an agent released from a source into the atmosphere or the sea, and its whirls and eddies carry the dispersed agent away from the source. When the location of the source is unknown, turbulence also obfuscates the ability of a remote sensor to infer the source location. This project exploits big-data strategies to develop new search algorithms that can identify the source of pollutants or other agents released in turbulent air or in the sea, using stationary or moving sensors. The principal objective is to directly enhance our readiness in responding to adverse events where the release of an agent threatens human or environmental safety. The ability to identify a contaminant source and dispersion pattern in complex turbulent flows has significant environmental and security implications. In cases where the strength of the source is not known, it must be inferred from remote measurements. When both the location and strength of the contaminant source are unknown, tracking the scalar source is an ill-posed problem and, as a result, most challenging. These tasks are often also the most urgent in responding to environmental and security threats. Preparedness in such circumstances requires prior planning and adaptive real-time capabilities. Therefore, both optimal placement of stationary sensors and olfactory search algorithms are devised in order to accurately and efficiently identify the location and strength of contaminants - a formidable engineering challenge due to the highly intermittent signal at the detector and the complexity of the precursor turbulent dispersion path. Optimal strategies to decode that information are formulated using a big-data computational laboratory where the entire space-time evolution of various turbulent flows, computed from high-fidelity direct numerical simulations, can be probed in real-time. The source-identification algorithm is formulated as a variational problem where properties of the adjoint field are exploited to identify the source location. The variational approach provides a natural framework to optimize the placement of stationary sensors in a region of interest, and to guide the motion of an olfactory robot towards the source. Repeated virtual experiments using the publicly-accessible computational laboratory, the various flow configurations, and different levels of fidelity of the flow field provide an accurate assessment of the performance of the search strategy in presence of scale-dependent model uncertainty. Real-world performance of the search strategy is tested using an autonomous underwater vehicle.