Composite index development via data fusion for system monitoring, diagnosis and decision-making
MetadataShow full item record
In a modern multistage manufacturing system, numerous sensors are installed in each manufacturing device, equipment, or station for in-situ, real-time monitoring of the process variables and product quality. Though each sensing data plays a central role in the given task, it is desirable to build a composite index to assess the overall performance of the system based on associated sensing information. This thesis proposes methodologies to establish a composite index via data fusion for monitoring, diagnosis, and maintenance decision making in aspects of complex multistage manufacturing systems, which includes (1) A novel composite index to monitor real-time product quality for continuous production of Carbon Nanotube buckypaper. Massive high-dimensional data are collected from the Raman spectroscopy sensor. The obtained spectrum data provide detailed nanostructure information within seconds, which can be decomposed into multiple effects associated with diverse quality characteristics of buckypaper. However, the decomposed effects are still high-dimensional, and a systematic quantification method for buckypaper quality assessment has been lacking. The proposed construction scheme for the composite index integrates penalized mixed-effects decomposition (PMD), weighted cross-correlation, and maximum margin clustering methods, to deliver a single quality index for fast product quality monitoring. (2) A novel automatic analytical framework to streamline the identification procedure of the process variables influencing the quality variable. It is desirable to conduct data analysis to identify the dependency between process variables and the quality variables without the involvement of highly trained data scientists. When coping with potential quality problems of the hot-rolling process, it can automatically identify the prominent process variables. The proposed framework identifies the process variables affecting product quality through feature extraction, clustering, and running permutation tests on multiple types of statistical models. (3) A composite index, termed AvaIlability-Degradation (AID), is proposed for integrating the degradation status of sensing information at the component-levels with the machine availability information at the plant-level. A proactive maintenance plan gives a manufacturer the capability to prolong the life of machinery and prevent the unexpected breakdown of production. However, there are no systematic methodologies to implement the concept in a semiconductor manufacturing system with cluster tools. Towards this end, we have further developed a model for AID integrated optimal proactive maintenance scheduling, which incorporates prognostic and diagnostics information in a three-level proactive maintenance decision-making framework. Additionally, we have validated the proposed strategy with a simulated semiconductor manufacturing process.