CAREER: Physics-Reinforced Data-Driven Prognostics and Co-Design for Marine Hydrokinetic Energy Systems

该学院早期职业发展（CAREER）项目将从根本上推进与海洋和流体动力（MHK）能源系统（包括海流涡轮机和波浪能转换器）的监测和设计相关的知识。MHK系统可以为能源经济的多元化做出重大贡献，提高国家的能源安全并减少对化石燃料的依赖。然而，这些系统从强力资源（例如强水流和/或大浪）产生电力，其对设备施加比类似额定功率的风力涡轮机大几倍的物理应力。这些限制导致严格的设计要求，增加了资本成本。此外，操作和维护成本很高，因为由于其海上地理位置和恶劣的腐蚀性环境，设备的使用受到限制。本研究计画将提供理论与计算基础，以提升MHK系统的可维护性、生存性与效率。该项目的长期目标是将传统的MHK涡轮机设计过程从顺序方法（子系统单独设计，忽略子系统之间的强耦合，通常导致次优设计）转变为新型协同设计框架（同时考虑具有耦合子系统的整个MHK系统的控制、可靠性和运营支出）。这种在最早阶段的同步协同设计允许互利的子系统，并可以显着提高整个系统的性能。因此，该项目将改善能源系统，加快蓝色经济的进展。将与国家可再生能源实验室和行业合作伙伴合作，并通过开源工具传播成果，加速技术转让。研究成果将被纳入新的研究密集型课程和新的能源弹性证书，并将为来自STEM代表性不足的群体的学生提供参与海洋可再生能源研究的机会。（剩余使用寿命的预测）和诊断MHK涡轮机（故障检测和识别）工具，目的是建立统一的设计框架，该框架考虑MHK系统的控制、可靠性和操作支出。该项目将探索从领域机械模型到深度学习的一系列工具。研究活动将以协同方式整合特定领域的物理知识和多源数据。具体而言，该项目将解决以下三个研究挑战。(1)数据稀缺的挑战：没有足够的数据来为MHK训练有效的诊断/诊断模型，因为该行业是新的。该项目将开发一种新的物理学强化知识转移学习方法，用于设计高效的模型，该模型使用风大数据作为源域，受MHK中的物理学约束作为目标域。(2)数据质量和概念漂移挑战：系统动态可能会随着时间的推移而变化，传感器数据可能会出现故障，因为MHK设备将部署在恶劣的偏远地区进行长期运行。本计画将发展一种新的图与强化学习方法，利用感测器网路结构资讯与多感测器时间序列的串流模式来设计强健的模型。(3)异构、多向耦合和协同优化挑战：应同时设计涡轮机几何形状、控制、可靠性和维护策略，以优化MHK涡轮机性能。本计画将以实验数据与动态模拟为基础，建立反应面模型，以反映设计参数与效能指标间的关系。将开发基于深度神经决策树的白盒协同优化工具，以优化涡轮机设计参数。总之，该研究成果将为陆上和海上浮动风电场、互联车辆和智能结构等复杂大型动态系统的稳健预测监测和协同设计奠定坚实的基础。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This Faculty Early Career Development (CAREER) project will fundamentally advance knowledge related to the monitoring and design of marine and hydrokinetic (MHK) energy systems, including marine current turbines and wave energy converters. MHK systems could contribute significantly to a diversified energy economy, improving the nation’s energy security and reducing reliance on fossil fuels. However, these systems generate power from puissant resources, such as strong water currents and/or large waves, which impose physical stresses on the equipment that are several times greater than wind turbines of similar power ratings. These constraints lead to stringent design requirements that increase capital costs. Further, operation and maintenance costs are high because access to equipment is limited due to their offshore geographical location and harsh corrosive environments. This research project will provide the theoretical and computational foundation to enhance MHK systems’ maintainability, survivability, and efficiency. The long-term goal of this project is to transform the conventional MHK turbine design process from a sequential approach, where subsystems are designed individually and strong coupling among them is neglected, generally leading to a suboptimal design, to a novel co-design framework that simultaneously accounts for control, reliability and operational expenditure of the overall MHK system with coupled subsystems. This simultaneous co-design at the earliest stage allows for mutually beneficial subsystems and could significantly improve the overall system performance. This project will thus improve energy systems and accelerate progress in the blue economy. Results will be disseminated in collaboration with the National Renewable Energy Lab and industry partners, as well as through open-source tools, accelerating technology transfer. Outcomes will be integrated into new research-intensive curricula and a new energy resiliency certificate, and opportunities will be provided to students from groups underrepresented in STEM to participate in marine renewable energy research.This research project aims to develop efficient and robust prognostics (prediction of the remaining useful life) and diagnostics (fault detection and identification) tools of MHK turbines, for the goal of establishing a unified design framework that accounts for control, reliability, and operational expenditure of MHK systems. The project will explore a spectrum of tools from domain mechanistic models to deep learning. The research activities will integrate domain-specific physics knowledge and multi-source data in a synergistic manner. Specifically, the project will address the following three research challenges. (1) The data scarcity challenge: there are not enough data to train an effective prognostics/diagnostics model for MHK because the industry is new. This project will develop a novel physics-reinforced knowledge transfer learning approach for designing efficient models that uses wind big data as the source domain, constrained by the physics in MHK as the target domain. (2) The data quality and concept drift challenge: system dynamics may change over time and sensor data are subject to failures because MHK devices are to be deployed in harsh, remote areas for long-term operation. This project will develop a novel graph and reinforcement learning approach for designing robust models using both sensor network structure information and stream pattern of multi-sensor time series. (3) The heterogeneous, multi-directional couplings and co-optimization challenge: turbine geometry, control, reliability, and maintenance strategies should be designed simultaneously to optimize MHK turbine performance. This project will build responsive surface models, that represent the relationships between design parameters and performance index, based on both experimental data and dynamical simulations. White-box co-optimization tools based on deep neural decision trees will be developed to optimize the turbine design parameters. Taken together, results from this research will establish a solid foundation for robust predictive monitoring and co-design of complex large-scale dynamic systems, such as onshore and floating offshore wind farms, connected vehicles, and intelligent structures.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.