Regression-based Quality Improvement in COmplex Systems with COnsideration of Data Uncertainty

Rapid advances in sensors and distributed sensing technologies have resulted in datarich environments, creating unprecedented opportunities for quality improvement in many domains. To make full use of the data for quality improvement, various statistical and data mining methods have been developed for process monitoring, abnormality detection, and fault diagnosis. However, most of these methods reply on a common assumption that the measured values of variables are the true values, with limited consideration of various types of uncertainty embedded in the measurement data. On the other hand, research on measurement errors has been conducted from a pure theoretical statistics point of view; that is, it focuses on studying impacts of measurement errors on statistical inference, without linking modeling and analysis of measurement errors with quality improvement objectives. Therefore, the objective of this project is to develop new methods for quality improvement in data-rich complex systems/domains by considering various types of data uncertainties. The data uncertainty defined in this project is different from sampling errors caused by limited sample size. Rather, it refers to those errors introduced into the data by physical devices or human factors during the process of data collection, storage, and retrieval, which leads to measurements of variables not exactly reflecting the variables true states. This project includes four inter-related research tasks: (1) process modeling, monitoring, and fault detection based on data with uncertainty; (2) separation of process fault and sensor fault; (3) optimal sensor allocation for fulfilling prescribed requirements on monitoring and diagnosis capabilities, using as few as possible number of sensors; and (4) identification of the maximum allowable level of data uncertainty under prescribed requirements on monitoring and diagnosis capabilities. In addition, this research will be validated and demonstrated in a multistage complex manufacturing process as well as in the domain of Alzheimers disease management.