Towards the integration of data-dependent process monitoring tasks using a time-frequency representation

Efficient process operation requires integration of various operation tasks and the techniques used to solve them. In this paper the author presents a formal framework for the integration of various data analysis and information extraction tasks. The author shows that a time-frequency representation of process data obtained via wavelet packet analysis, provides a consistent framework for unifying the tasks of outlier detection, constraint-free dynamic data reconciliation, and univariate data compression. Furthermore, the resulting data representation in the time-frequency domain may be exploited very effectively for information extraction via multivariate statistical methods. The data representation obtained after rectification can be seamlessly integrated with existing algorithms for fast matrix multiplication and adaptive approximate principal component analysis