Data fusion methods for modeling and monitoring in complex systems

Abstract

Recent advances in measurement and sensing technology have generated data-rich environments in many industrial and service applications with complex systems, which has led to an increasing need to handling datasets generated from multiple sources. These data contain detailed information of the engineering process, and provide great opportunity for getting better understanding of the system and thus improving quality. However, difficulties and challenges remain in modeling and monitoring data from such systems due to great variety and variability of such datasets with various sources. This thesis contains three projects which conducts data fusion methods to address the issues in different applications. In the first project, a hierarchical Bayesian method is proposed to model and monitor the customer reviews with both textual content and numerical ratings in E-commerce feedback systems. The second project is concerned with modeling and improving estimation of covariance structure of data from multiple sensors with combining associated data sources including geographical information and group information. The method is applied in a sensor system designed to detect and predict landslides and hill-slopes. The third project propose a statistical transfer learning based method to integrate information from multiple sites and sensors with auto-correlated sensor readings for newly set-up sensors. These essays provide effective solutions of integrating multiple data sources to help modeling and monitoring complex systems appeared in modern engineering applications.

Date of Award	2018
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology