Causation-based quality control methodologies with applications

Distributed sensing networks (DSN), a system-wide deployment of different types of sensing devices in manufacturing systems, have resulted in a data-rich environment that is both temporally and spatially dense, which provides unprecedented opportunities as well as challenges for quality improvement. Existing quality control techniques fail to fully utilize the distributed sensing data, because most techniques were not purposely developed for analyzing these datasets which are massive, high-dimensional, heterogeneous, and contain substantial uncertainty. Therefore, it is urgent and essential to develop new methodologies to make effective use of the data for quality control.; This dissertation research aims to develop "causation-based quality control" methodologies in order to establish a new science base with a set of tools for "causation-based" process monitoring, diagnosis, and control, which provides a highly efficient and reliable means for manufacturing quality improvement in this data-rich environment.; This dissertation consists of four major chapters. Chapter 2 investigates causal modeling in a multistage rolling process by integrating manufacturing domain knowledge with statistical data analysis. This leads to an efficient and effective algorithm to identify causal relationships from production data and enables quality control of the process. Chapter 3 investigates causation-based monitoring and diagnosis by using the causal relationships to guide the decomposition procedure in a traditional diagnostic approach. This avoids excessive computational efforts and results in significant enhancement of diagnostic accuracy. Chapter 4 investigates the robustness of the causal modeling with respect to data uncertainty by analytically deriving an upper bound of the allowable uncertainty. This addresses the challenges of the causal modeling in the DSN environment and provides solutions to a problem not well studied in causal modeling literatures. Chapter 5 extends the causal modeling research to an epidemiologic application by identifying the factors that influence disease contraction. This provides an effective way to analyze heterogeneous public health databases for knowledge discovery and supporting the decision making in disease control and diagnosis.; This research is the first effort that introduces causal modeling and analysis into the quality engineering discipline. It provides enabling methodologies and algorithms to address the challenges arising in this data-rich era for quality control and improvement.