Decentralized Data Analytics and Optimization Methods for Physical Asset Management

现有的分布式物理资产网络的数据驱动预测模型是集中的，依赖于资产传感器、故障和异常数据的可用性。数据稀缺的问题通常是通过聚集来自类似资产的数据来创建更大的诊断数据池来解决的。但是，如果相似的资产属于不同的企业，它们可能不愿意彼此共享其原始资产数据，或者将其发送到中央服务器进行处理。此外，现有的维护和任务分配调度优化模型是集中的。这些模型没有考虑过程中涉及的各种涉众的目标和偏好，并且需要访问涉众的私有信息，例如他们的生产计划和资产的状态/可用性。然而，云制造、智能电网、网联汽车和医疗保健中使用的许多物联网（IoT）资产和网络物理系统（CPS）在地理上是分散的，由不同的企业拥有，这使得现有模型不再适用或适合这些资产。考虑到最近技术进步在提供实时连接、数据处理和信息共享方面的潜力，本研究旨在为分布式物联网资产网络的预测、维护调度优化、实时任务分配和调度优化开发新的分散解决方案。拟议的研究计划将推进云边缘基础设施中分布式物联网资产管理的知识。分散的预测模型将通过在全局模型中聚合来自多个组织的贡献来提高故障预测的准确性，同时允许组织保持其数据的私密性。这种更准确的故障预测将大大降低维修成本，减少故障对环境、经济和社会的负面影响。分散的维护、任务分配和调度优化模型将使优化过程具有层次化、协同化的知识创造、更好的资源利用、更高的能源效率、更强的弹性和更短的计算时间。在实现这些结果的同时，组织仍然可以在不完全共享其私人数据的情况下最大化自己的利益。提出的模型将帮助组织过渡到充分实现和利用新兴技术，如云和边缘计算。高素质人才（HQP）将在数据分析，数学建模和优化方面拥有独特的技能，这将为他们在学术界或工业界创造特殊的职业机会。

Nature of the Proposed Research The existing data-driven prognostics models for a network of distributed physical assets are centralized and reliant on the availability of assets' sensors, failures and anomaly data. The issue of data scarcity is usually tackled by aggregation of data from similar assets to create a larger data pool for diagnosis. However, if the similar assets belong to various enterprises, they may be reluctant to share their raw asset data with each other, or send it to a central server for processing. In addition, the current optimization models for maintenance and task allocation and scheduling are centralized. These models do not consider the objectives and preferences of various stakeholders involved in the process, and require access to stakeholders' private information, such as their production schedules and the status/availability of their assets. However, many Internet of Things (IoT) assets and cyber-physical systems (CPS) used in cloud manufacturing, smart grids, connected vehicles, and healthcare are geographically distributed and owned by different enterprises, which make the existing models no longer applicable or appropriate for these assets. Realizing the potential of recent technological advancements in providing real-time connectivity, data processing, and information sharing, the proposed research aims to develop novel decentralized solutions for prognostics, maintenance scheduling optimization, and real-time task allocation and scheduling optimization of a network of distributed IoT assets. Anticipated Outcomes The proposed research program will advance knowledge for distributed IoT asset management in Cloud-Edge infrastructures. The decentralized prognostic models will enhance failure prediction accuracy by aggregating the contributions from multiple organizations in a global model while allowing the organizations to keep their data private. This more accurate failure prediction will significantly lower maintenance costs, and decrease the negative impacts of failures on the environment, economy and society. The decentralized optimization models for maintenance, task allocation, and scheduling will result in hierarchical and collaborative knowledge creation, better utilization of resources, higher energy efficiency, more resilience, and lower computation time of the optimization process. These outcomes will be achieved while organizations still can maximize their own benefits without fully sharing their private data. The proposed models will help organizations in transition towards full realization and utilization of emerging technologies, such as cloud and edge computing. Highly qualified personnel (HQP) trained in this program will possess a unique skillset in data analytics, mathematical modeling and optimization which will create exceptional career opportunities for them in academia or industry.