A hierarchical approach to data driven fault detection and diagnosis (FDD)

过程/设备性能管理被认为是一项具有挑战性的任务，因为我们处理的过程大多是动态的，本质上是非线性的，并且相互关联。通过简单的过程数据趋势分析，不容易检测和识别基础变量之间关系的初期变化。对于系统/过程，故障检测和诊断（FDD）应该在不同的级别（例如传感器级、设备级、单元级等）进行，由于在每个层次上，可用于进行FDD的信息和资源是不同的，并且FDD的算法和设计目标通常也是不同的。显然，对于一个真正自主和智能的健康监测系统，需要开发一个元策略，以集成不同的设计方案，基于特定的设计目标，在不同的资产层次。此外，数据驱动FDD在设备级和单元级的应用被认为是新颖的。在这个Engage Plus Grant项目中，我们的目标是与霍尼韦尔过程解决方案密切合作，通过应用分层深度学习方法来研究和开发创新的数据驱动的故障检测和诊断框架。

Process/Equipment performance management is considered to be a challenging task because processes we aredealing with are mostly dynamic, nonlinear in nature and interconnected. Incipient changes in relationshipbetween underlying variables are not easy to detect and identify through simple process data trending. For asystem/process, fault detection and diagnosis (FDD) should be done at different levels (e.g. sensor level,equipment level, unit level, etc.), since at each level the information and resources available for conductingFDD are different, and algorithms and design objectives of FDD are usually different too. Clearly for a trulyautonomous and intelligent health monitoring system, a metastrategy needs to be developed to integratedifferent design schemes based on specific design objectives at different asset hierarchy. Furthermore,application of data driven FDD at both equipment and unit-level is deemed novel. In this Engage Plus Grantproject, we aim to work closely with Honeywell Process Solutions to investigate and develop an innovativedata-driven fault detection and diagnosis framework by applying a hierarchical deep learning approach.