GOALI: Processing, System Modeling and Process Control for Complicated Functional Data

该奖项为学术界与工业界的联系提供了资助机会，用于开发用于处理制造系统中具有急剧变化和非平稳模式的大尺寸功能数据的方法和算法。 所开发的方法将整合基于小波的信号处理技术和面向神经网络的系统建模程序，以开发用于超薄半导体薄膜沉积等制造过程的原位过程控制工具。 该研究将优化一个通用的数据减少目标函数，平衡数据压缩目标和各种类型的数据分析和决策使用的建模精度要求。 所开发的统计过程控制算法，旨在研究过程偏离选择小波参数将更有效地检测局部变化，监测复杂的功能数据。 通过模拟不同数量的数据噪声和信号变化模式，从文献中获得的各种测试数据曲线上的实验开发的程序将提供对扩展现有工作的数据减少的目的，所提出的方法的优点和缺点的见解。 我们的行业合作伙伴PDF Solutions，Inc.将与我们合作，为调整所研究的方法解决实际问题提供见解，并在实际应用中测试所开发的方法和工具。如果成功，本研究的结果将导致现场过程控制和局部聚焦神经网络算法的改进，以及大规模复杂功能数据的分析和建模的新发展。 这项工作的主要目标是提供一个通用的数据减少工具，在挖掘知识，从大量和复杂的功能数据的广泛应用。 获得相关知识将有助于从业人员更有效地识别过程问题，更有效地提高制造系统的质量。 在这个跨学科项目中学到的经验将滋养我们的综合教育和研究计划，以培养下一代制造企业系统工程师。

This Grant Opportunity for Academic Liaison with Industry (GOALI) award provides funding for the development of methodology and algorithm for processing large-size functional data with sharp changes and nonstationary patterns from manufacturing systems. The developed methods will integrate wavelet-based signal-processing techniques and neural-networks oriented system-modeling procedures to develop an in-situ process control tool for manufacturing processes such as ultra-thin semiconductor-film deposition. The research will optimize a generic data-reduction objective function balancing data-compression goals and modeling accuracy requirements for various types of data analysis and decision-making use. The developed statistical process control algorithm aiming to study process deviates at selective wavelet parameters will be more effective in detecting local changes in monitoring complicated functional data. Experimenting the developed procedures on various testing data curves obtained from the literature by simulating different amount of data noises and signal-change patterns will provide insights of the strength and weakness of the proposed methods against extensions of existing work for data-reduction purposes. Our industry partner from PDF Solutions, Inc. will work with us in providing insights for tuning the researched methods in solving real-life problems and in testing the developed methods and tools in practical applications.If successful, the results of this research will lead to improvement in algorithms for in-situ process control and a locally focused neural network, and new development in analysis and modeling of large-size complicated functional data. The primary goal of this work is to provide a generic data-reduction tool with a wide range of applications in mining knowledge from massive and complex functional data. Getting the relevant knowledge will help practitioners identify process problems more efficiently and improve the quality of manufacturing systems more effectively. The experience learned in this interdisciplinary project will nourish our integrated education and research program for training the next generation of manufacturing enterprise system engineers.