Development of Data-driven Decision Support System using Deep Learning Techniques

这一提议的动机是需要开发一个数据驱动的决策支持系统框架，该框架可用于两个应用领域，即航空航天系统的预测和健康管理以及森林管理服务。这项合作研究的主要目标是开发一个决策支持框架，该框架使用基于高级深度学习(DL)的算法按顺序执行数据处理、变化/异常检测、分类、预测分析和决策支持，以帮助PHM和森林管理应用领域的决策者。该项目涉及两个行业合作伙伴，Tecsis Corporation和Hegyi Geomatics Inc.。Tecsis将为其PHM应用程序使用拟议的框架，为用户提供感知组件或子系统的健康状态并预测其未来维护策略的能力。Hegyi将采用其森林管理应用程序的拟议框架来获取和分析森林信息，以帮助管理者优化森林资源的整体效益。在主题1中，将使用混合公式通过利用每种方法的优点来研究各种DL方法对多类分类的增强。此外，与数字学习方法相关的一些挑战将通过探索迁移学习、领域适应和数据增强技术来解决。主题2将集中于剩余的可用寿命预测和生物量估计，通过探索各种DL方法在处理时间序列数据和图像数据方面的优势。为了进一步提高这些动态学习方法在精度和稳健性方面的性能，将研究领域知晓的动态学习方法、集成学习技术和不确定性量化方法。主题3将侧重于根据通过分类和预测建模方法积累和理解的信息收益开发决策支持模块，这些方法得到拟议的分布式决策支持系统框架中包括的深度学习技术的支持。将使用Tecsis和Hegyi为其各自的应用提供的数据集，展示拟议的DDSS框架的有效性和优势。三名硕士、四名本科生、一名博士和一名PDF将在即将到来的数据分析和深度学习领域接受该项目的培训。通过采用这项研究的结果，Tecsis和Hegyi希望扩大他们在各自行业的研发服务和客户基础。

This proposal is motivated by the need to develop a data-driven decision support system (DDSS) framework that can be used in two application areas, namely, prognostics and health management (PHM) of aerospace systems, and forest management services. The main goal of this collaborative research is to develop a decision support framework that uses advanced deep learning (DL) based algorithms to perform data processing, change/anomaly detection, classification, predictive analytics, and decision support in a sequential order to aid decision makers in the applications areas of PHM and forest management. There are two industrial partners involved in this project, Tecsis Corporation and Hegyi Geomatics Inc. Tecsis will use the proposed framework for their PHM applications to provide users with the ability to perceive the health state of a component or subsystem, and to predict its future maintenance policies. Hegyi will adopt the proposed framework for their forest management applications to acquire and analyze forest-based information to assist managers in optimizing the overall benefits of forest resources.In this proposal we have three research themes. In Theme 1, enhancements to multi-class classification will be investigated using hybrid formulations various DL methods by exploiting the strengths each method. Further, some of the challenges associated with DL methods will be addressed by exploring transfer learning, domain adaptation, and data augmentation techniques. Theme 2 will focus on the remaining useful life prediction and biomass estimation by exploring the strengths various DL methods in handling time series data and imagery data. To further improve the performance of these DL methods in terms of accuracy and robustness, domain-informed DL methods, ensemble learning techniques, and uncertainty quantification methods will be investigated. Theme 3 will focus on the development of decision support module based on the accumulation and understanding information gain through classification and predictive modeling methods supported by deep learning techniques included in the proposed DDSS framework. The effectiveness and advantages of the proposed DDSS framework will be demonstrated using the data sets provided by Tecsis and Hegyi for their respective applications. Three Master's, four undergraduates, one PhD and one PDF will be trained in this project in the upcoming areas of data analytics and deep learning. By adopting the results from this research, Tecsis and Hegyi expect to expand their R&D services and cliental base in their respective sectors.