Data Analytics and Learning Based Industrial Diagnostics and Monitoring Systems

在Engage Grant的支持下，与加拿大埃德蒙顿的Honeywell Connected Plant（HCP，前身为Honeywell Process Solutions）开展了合作研究，从2016年夏季开始，并在Engage Plus Grant的支持下于2017年继续进行。自主数据驱动的故障诊断（FD）和监测计划已经开发。基于核心数据驱动模型构建的是一个用于自主检测、诊断和预测的分层框架。在复杂的工业过程中，采集的数据具有高维、非线性、时变性和自相关等特点。因此，来自先前合作的挑战之一涉及数据模型的准确性，也称为（实际过程/工厂的）“数字孪生”。此外，基于KPI预测的预测也很有意义。 这促使我们研究最新的人工智能风味的机器学习技术，这些技术能够自动调整并且更加通用。在这个项目中，我们建议探索各种机器学习和人工智能技术，并将其部署到过程健康监控中。这将导致为行业合作伙伴开发的自主故障诊断和监控系统的第一个原型的显着改进。

Collaborative research has been carried out with Honeywell Connected Plant (HCP, formerly known as Honeywell Process Solutions), Edmonton, Canada, starting summer 2016, with the support of Engage Grant, and continued in 2017 with the Engage Plus Grant. An autonomous data-driven fault diagnosis (FD) and monitoring scheme has been developed. Built upon the core data-driven model is a hierarchical framework for autonomous detection, diagnosis and prediction. In complex industrial process, data collected demonstrates unique characteristics, including high-dimensionality, nonlinearity, time-variance, and auto-correlation. As a result, one of the challenges from the previous collaboration concerns accuracy of the data model, also referred to as "digital-twin" (of the actual process/plant). In addition, prognosis based on KPI prediction is of great interests. This motives us to investigate the most recent artificial intelligence flavoured machine learning techniques, which are capable of auto-tuning and are more versatile. In this project, we propose to explore various machine learning and AI techniques, and deploy them to process health monitoring. This will lead to a significant improvement of the first prototype of the autonomous fault diagnosis and monitoring system developed for the industry partner.