RECURRENCE ANALYSIS OF HIGH-DIMENSIONAL COMPLEX SYSTEMS WITH APPLICATIONS IN HEALTHCARE AND MANUFACTURING
Open Access
- Author:
- Chen, Cheng Bang
- Graduate Program:
- Industrial Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- July 22, 2019
- Committee Members:
- Soundar Rajan Tirupatikumara, Dissertation Advisor/Co-Advisor
Soundar Rajan Tirupatikumara, Committee Chair/Co-Chair
Saurabh Basu, Committee Member
Bharath Kumar Sriperumbudur, Outside Member
Hui Yang, Dissertation Advisor/Co-Advisor
Hui Yang, Committee Chair/Co-Chair
Ling Rothrock, Program Head/Chair - Keywords:
- Recurrence Analysis
Nonlinear Dynamics
Pattern-Frequency Tree
Network Science
Space-Filling Curve
Complex system
Healthcare
Manufacturing - Abstract:
- The rapid development of advanced sensing technologies yields a large volume of data that contain rich information about the systems. In general, real-world systems tend to be complex and need advanced methods to derive inferences. The fusion of information from diverse sensors and with different physical characteristics enhances the understanding of system principles and provides the fundamentals for planning, decision-making, and optimization. However, the proliferation of nonlinear, nonstationary, and high-dimensional data with complex structures impedes knowledge extraction from data, which brings significant challenges for human experts who use the collected data to inspect the integrity and performance of complex systems. Therefore, this dissertation attempts to develop innovative methodologies and frameworks addressing the characteristics of recurrence dynamics of high-dimensional data in optimizing decision-making and system performance. Specifically, this dissertation focuses on mining dynamic recurrences of complex systems in both temporal and spatial domains for feature extraction, system monitoring, and anomaly detection. Note that this dissertation will enable and assist in (i) sensor-driven modeling, monitoring, and optimization of complex systems, (ii) integrating design for complex nonlinear systems, and (iii) creating better predictive/diagnostics tools for real-world applications. The accomplishments include the following: Recurrence Pattern-Frequency Tree: In Chapter 2, a novel Pattern-Frequency Tree (PFT) approach is developed for multi-sensor signal fusion and dynamic transition analysis in high-dimensional environments. The PFT model characterizes both transition patterns and frequency information of high-dimensional dynamic transitions, and it facilitates analyzing and capturing the abnormal transitions in nonstationary nonlinear systems. Recurrence Network Analytics: In Chapter 3, a generalized recurrence network scheme is proposed to categorize the spatial recurrence patterns and characteristics effectively. The proposed methodology, considering both observation similarity and geometric features, quantifies and characterizes the spatial recurrence through the characteristics of the network topology. This approach provides not only excellent visualization of the recurrence patterns but also produces precise quantitative measures of the spatial recurrence. Heterogeneous Recurrence Analysis of spatial data: In Chapter 4, the recurrence analysis of spatial data is extended to characterize the heterogeneous recurrence dynamics in the spatial domain. The proposed method first leverages the space-filling curve to tackle the high dimensionality issue, and then develops a nonlinear state-space representation to extract the recurrence dynamics in the spatial systems. Finally, the heterogeneous recurrences of the spatial data are characterized by the fractal characters of the transformed data utilizing the iterated function system transformation. Furthermore, new quantifiers for heterogeneous recurrence quantification analysis (HRQA) are developed to quantify the heterogeneous recurrences. Experimental results show that the proposed approach yields superior performance in the extraction of salient features to characterize recurrence dynamics in the spatial systems compared to the conventional methods.