GOALI: Bayesian Hierarchical Network based Computational Framework for Risk Tolerant Process Design

这项学术联络与工业（GOALI）奖的研究目标是为安全关键部件（如风力发电机的薄壁轴承）制造过程的风险承受设计创建一个计算框架。这些部件容易变形和过早失效，这对风车系统的设计提供了不希望的约束。本研究将开发一种方法，利用分层贝叶斯网络方法在计算设计框架中捕获制造过程的不确定性和故障诊断。它将从在概率公式中表示过程不确定性和相关的失效响应发展到将其嵌入到过程设计的连续表示及其在几何，微观结构和损坏状态中的相关转换中。材料状态的物理模型将包括过程方差与失效状态和风险的关系。交付成果包括流程设计的计算工具，风险贝叶斯推理算法，风力发电系统设计的应用结果，以及校准和验证实验的结果。如果成功，这项研究将使下一代产品的设计师在设计决策中考虑工艺设计和由此产生的性能不确定性。他们不仅可以提高现有设计的使用寿命和可靠性，还可以设计出更轻、功率密度更高、使用寿命更长的产品。示例应用包括大型薄壁风车和航空发动机轴承，发电传动部件和核工业安全关键部件。在这些地区，不良工艺设计的风险具有严重的社会后果。工程专业的学生，尤其是少数族裔和女性学生，将从研究生和本科课程中引入概率计算过程设计中受益匪浅。此外，学生将有机会与工业合作伙伴铁姆肯公司在项目团队中工作，并参加暑期实习项目。

The research objective of this Grant Opportunity for Academic Liaison with Industry (GOALI) award is to create a computational framework for risk tolerant design of processes in the manufacture of safety critical parts such as thin walled bearings for wind mills. These parts are prone to distortion and premature failure which provides an undesirable constraint on the design of windmill systems. This research will develop a method that captures manufacturing process uncertainties and failure diagnostics in the computational design framework using hierarchical Bayesian network approach. It will progress from the representation of process uncertainties and associated failure response in a probabilistic formulation to embedding this in the continuum representation of process design and its associated transformations in geometry, microstructure and damage state. The physical models of the material state will include the relationship of process variance with the failure state and risk. Deliverables include computational tools for process design, algorithms for Bayesian inference of risk, results of application to design of wind mill systems, and results of calibration and validation experiments.If successful, this research will enable designers of next generation products to consider process design and resulting performance uncertainty in their design decisions. They will be able to not only increase the service life and reliability of their current designs but also design products of which are lighter, with higher power density and longer service lives. Example applications include large thin walled windmill and aeroengine bearings, transmission components in power generation and safety critical parts in nuclear industry. In these parts the risk of poor process design has severe societal consequences. Engineering students especially minorities and women will benefit greatly from the introduction of probabilistic-computational process design in graduate and undergraduate curricula. In addition, students will get opportunities to work with the industrial collaborator Timken in project teams as well as in summer internship programs.