| Adaptive and automated multiscale data analysis seeks to extract relevant information from data without imposing too many restrictions, allowing the data to speak for itself. Multiscale data analysis is essential when the system (or data) understudy is influenced by mechanisms whose interactions produce uncertain behavioral patterns in the observed data. Also, when a priori knowledge of the system is not clear or when the observed phenomenon is non-stationary, a multiscale data analysis approach could be helpful. Continuous monitoring and evaluation of systems requires the acquisition of data which are mostly inherently non-stationary. By taking advantage of recent advances in adaptive data analysis, tremendous improvements could be made in areas such as automation, feature extraction and tracking, and modeling and forecasting for data analysis systems. This dissertation aims to improve the performance of a recently developed adaptive signal analysis algorithm, Empirical Mode Decomposition (EMD) and its applicability to real world datasets. Contributions made in this direction are three-fold: firstly, a new methodology for reducing end-point interpolation problems through end-point stabilization is introduced. Also, algorithms for the reconstruction of decomposition results are introduced. This is essential for understanding underlying mechanisms generating observed data as it enhances the physical meaning of the Intrinsic Mode Functions (IMFs). Second, a parallel implementation of the 2-dimensional extension of the EMD has improved on the practical application of the algorithm to different datasets due to speed up improvements achieved. Lastly, the algorithm is successfully integrated in a Geographic Information Systems (GIS) platform for pavement condition evaluation, visualization and prioritization. Substantial contributions have also been made in the area of pavement crack edge detection by the use of computer vision tools, which endow computers with information-processing capabilities. This enables engineers and scientists to model and automate the process of visual recognition in a way comparable to those of biological organisms. Finally, successful application of this work in three different civil engineering research areas: oceanography, traffic engineering and pavement management, is presented through the various case studies undertaken. |
