Event Detection Platform (using pattern matching analytics)

这个为期18个月的项目旨在建立一个由IT平台组成的商业前演示，该平台能够部署CyBula的事件检测方法来监控复杂资产(机器、发动机、基础设施和医疗应用、人员)的行为。这些方法对多变量、时间序列数据使用新的模式匹配技术；软件分析非常受欢迎，因为许多组织被监控的资产数据淹没，而他们缺乏与数据交互的工具。这些分析帮助回答的基本问题是：(I)资产是否按预期运作？(二)如有异常，该事件过去曾否发生？(Iii)从以前的事件中可以得出什么结论，最后(Iv)在需要采取行动之前，我们还有什么时间？CyBula的Signal Data Explorer是一个台式软件包，专家可以使用它来浏览大量的多变量数据，然后建立异常模型或事件签名，然后在存档的数据上验证它们。这种测试环境吸引了几个市场应用(航空航天、铁路、能源生产、石油和天然气生产以及重症监护医疗)中的几家“蓝筹”公司的参与，CyBula继续投资于新功能。该演示程序将允许这些模型和事件签名作为工作流从Signal Data Explorer导入，并部署在连续的流传输资产数据上。用户将能够构造他们的资产数据并附加输入参数以馈送不同的模型。可视化模块将使用户能够配置屏幕显示以可视化随着时间的推移检测到的异常或事件。该项目预计该演示平台能够处理涉及数千个被监测变量和多达100个检测模型的大规模资产监测问题。

This 18 month project aims to build a pre-commercial demonstrator consisting of an ITplatform which is capable of deploying Cybula’s event detection methods to monitor thebehaviour of complex assets (machines, engines, infrastructure and with medical applications,people). These methods use novel pattern matching techniques on multi-variate, time-seriesdata; software analytics that are very much in demand as many organisations are swampedwith the flow of monitored asset data and they lack the tools to interact with the data. Thefundamental questions which these analytics help to answer are (i) is the asset operating asexpected? (ii) if there is an abnormality, has this event occurred in the past? (iii) what can belearnt from previous events and finally (iv) what time do we have before action is required?Cybula’s Signal Data Explorer, a benchtop software package, can be used by specialists tobrowse large volumes of multi-variate data before building abnormality models or eventsignatures and then validating them on archived data. This testing environment has attractedthe involvement of several ‘blue-chip’ companies across several market applications(aerospace, rail, energy generation, oil & gas production and critical-care medicine) andCybula continues to invest in new features.This demonstrator will allow these models and event signatures to be imported as workflowsfrom Signal Data Explorer and deployed on continuously streamed asset data. The user willbe able to structure their asset data and attach input parameters to feed the different models.A visualisation module will enable the user to configure screen displays to visualise detectedabnormalities or events over time. The project anticipates the demonstration platform to becapable of handling large scale asset monitoring problems involving thousands of monitoredvariables and up to 100 detection models running simultaneously.