GOALI: Causation-Based Quality Control - A New Paradigm to Achieve Effective Monitoring, Diagnosis, and Control for Complex Manufacturing Systems

这项赠款资助发展“基于因果关系”的质量控制方法。采用一般概率网络或概率图模型表示的因果模型来表示质量和过程变量之间的因果关系。 第一个努力是开发工程知识增强的因果建模，重点是将工程领域的知识整合到因果结构学习和参数估计中，并考虑传感数据的不确定性。 在因果模型的基础上，通过因果关系的显式分解，开发基于因果关系的监测和诊断算法，以提高诊断能力。 此外，在线基于原因的控制和干预将研究结合谨慎控制原则与因果模型，考虑模型参数和干预的不确定性。 最后，开发的方法将通过与工业公司的GOALI努力在多阶段成形过程中进行验证和实施。 工业企业将为研究团队提供真实的数据、案例和制造工艺信息。 国际合作研究和教育工作将在整个研究项目中进行。该项目的成功将通过为信息处理能力提供新的概念、标准和算法来提高质量，从而推进复杂系统建模和控制的最新技术水平。 该项目创建了使能方法，在数据丰富的环境中从传统的SPC过渡到主动、实时诊断和预测控制模式。 开发的基于因果关系的质量控制方法将提供算法和相关软件工具，以实现（i）基于因果关系（而不仅仅是相关性）的过程建模和分析;（ii）嵌入式诊断监控和根本原因识别（而不仅仅是变化检测）;以及（iii）在线质量推断和缺陷预防干预（而不仅仅是离线模拟和质量检查）。 所开发的方法解决了数据丰富的环境中的关键问题，这是一个挑战所有行业部门的问题。由此产生的方法的实施预计将产生广泛的经济影响。

This grant funds the development of "causation-based" quality control methodologies. A causal model, represented by a general probabilistic network or a probabilistic graphical model, is adopted to represent causal relationships among quality and process variables. The first effort is to develop engineering knowledge enhanced causal modeling with the focus on integrating the engineering domain knowledge into the causal structure learning and parameter estimation with the consideration of sensing data uncertainties. Based on the causal model, causation-based monitoring and diagnostic algorithms will be developed through explicit decomposition according to their causal relationships for diagnostic capability enhancement. In addition, on-line causation-based control and intervention will be studied by integrating cautious control principles with the causal model considering the uncertainties of both model parameters and intervention. Finally, the developed methodologies will be validated and implemented in multistage forming processes through a GOALI effort with industrial companies. Industrial companies will provide real data, cases, and manufacturing process information for the research team. International collaborative research and education efforts will be pursued throughout the research project.The success of this project will advance the state of the art in modeling and control of complex systems by contributing new concepts, criteria, and algorithms to the information-processing capabilities for quality improvement. The project creates enabling methodologies to bring a transition from a traditional SPC to a proactive, real-time diagnostic and predictive control paradigm in a data-rich environment. The developed causation-based quality control methodologies will provide algorithms and associated software tools to achieve (i) causation (rather than only correlation) based process modeling and analysis; (ii) embedded diagnostic monitoring and root cause identification (rather than only change detection); and (iii) on-line quality inference and intervention for defect prevention (rather than off-line simulation and quality inspection). The developed methodology addresses critical issues in a data rich environment, which is a problem challenging all industry sectors. The implementation of the resulting methodologies is expected to generate broad economic impacts.