CAREER: Process Monitoring, Identification, and Diagnosis Using Image and Functional Data

该CAREER项目将建立一个教育和研究计划，用于过程监控、故障识别和故障诊断的方法，使用高维功能数据或图像传感器信号。由于图像和功能数据量大、维数高，及时获取综合信息进行质量改进所需的分析和操作变得复杂且资源密集。所提出的方法将集成基于小波的信号处理技术和面向数据挖掘的系统建模程序，以开发制造过程的过程中（或原位）控制工具。提出的方法包括：(1)一种针对功能数据或图像传感器信号的自适应多尺度监测模型，以克服现有技术存在的故障检测概率低的问题；(2)基于核函数的故障识别指标度量和变量选择算法来识别贡献过程变量，提高诊断性能；(3)非线性故障诊断模型，以确定故障的可分配原因，并有效地处理病态自相关功能数据或图像传感器信号。这项研究的结果将为制造商提供处理高维功能数据或图像传感器信号的新工具，以提高其制造过程的质量。所提出的方法的优势在于它是通用的，可以适用于其他研究和应用领域，例如非线性剖面的监测、大规模传感器故障检测问题和光谱数据的变量选择问题，或者在生物过程、半导体和汽车制造以及其他制造过程中与非线性多元SPC相关的新研究问题的发展。

This CAREER project will build an education and research program in methodology for process monitoring, fault identification, and fault diagnosis using high-dimensional functional data or image sensor signals. Due to the massive amount and high dimensionality of image and functional data, the analysis and manipulation required to obtain timely synthesized information for quality improvement becomes complicated and resource intensive. The proposed methods will integrate wavelet-based signal-processing techniques and system-modeling procedures oriented to data mining to develop an in-process (or in-situ) control tool for manufacturing processes. The proposed methodology includes the following: (1) an adaptive multi-scale monitoring model for functional data or image sensor signals to overcome the problem of the low fault detection probability present in existing techniques; (2) a kernel-based fault identification index measure and a variable selection algorithm to identify contributing process variables to improve diagnostic performance; (3) a nonlinear fault diagnosis model to determine an assignable cause for a fault and to effectively handle ill-posed autocorrelated functional data or image sensor signals.The results of this research will provide manufacturers with new tools for handling high-dimensional functional data or image sensor signals for quality improvement of their manufacturing processes. The strength of the proposed methodology is that it is generic and can be adapted to other research and application areas, such as in the monitoring of nonlinear profiles, the problem of large-scale sensor failure detection, and variable selection problems of spectrum data, or in the development of new research issues related to nonlinear multivariate SPC in bioprocesses, semiconductor and automobile manufacturing, and other manufacturing processes.