Single Channel ICA And Its Application In Deformation Analysis

:Seperating noise, systematic error and deformation components caused by different forces or deformation information with different characteristics from deformation monitoring data is a main step for deformation analysis, which would contribute to engineering structural damage identification and deformation prediction. With the assumption that source signals are mutually independent, Independent Component Analysis (ICA), a blind signal separation method, can separate the source signals from their mixtures, which makes the ICA a powerful tool for deformation analysis. In addition to independence assumption, another requirement for ICA is that the number of mixtures should be no less than that of source signals. Unfortunately, only one channel of observed deformation data series is availbale for one monitoring point in most of the case. Therefore, the ICA with single channel data input (called single channel ICA) is applied into deformation analysis. Based on a comparative analysis of different single channel ICA algorithms, phase space reconstruction based single channel ICA is researched and essential empirical mode decomposition based single channel time-varying ICA is proposed for deformation monitoring data de-noising, deformation components decomposition and deformation modeling. The main researches are following:1. The basic theories of ICA are studied in details and several commonly used ICA algorithms such as the FastICA algorithm and multidimensional independent component analysis are summarized.2. A comparative analysis of dynamic embedding based single channel ICA (DE-ICA), wavelet transform based single channel ICA (WT-ICA), empirical mode decomposition based single channel ICA (EMD-ICA) and essential empirical mode decomposition based single channel ICA (EEMD-ICA) is made in the simulation experiments. The results indicate that DE-ICA is the optimal, EMD-ICA is simple without any additional parameters, both of WT-ICA and DE-ICA are fastest in the operating efficiency, EMD-ICA is second and EEMD-ICA is last due to several iterations.3. To solve redundancy problem caused by too many embedding dimensions in DE-ICA, phase space reconstruction based single channel ICA (PSR-ICA) is proposed to choose optimal embedding parameters. The result of simulation experiment shows that PSR-ICA is not only an efficient tool for data de-noising, but also able to separate independent components from observation data.4. PSR-ICA is applied to process Wuqiangxi Dam displacement monitoring data for deformation analysis. The PSR-ICA is conducted twice for measurement de-noising and displacement component extraction successively. The displacement compoments by the second PSR-ICA process were interpreted as function of the causative factors such as temperature and water level. The result demonstrates that the mainly two separated displacement components are almost consistent with not only the temperature and water level but also their displacement components computed using the regression model. Therefore it also verifies that PSR-ICA could separate the causative components from dam displacement efficiently.5. An online blind source separation ICA algorithm, called EASI algorithm, is researched. Unlike FastICA, it is intended to solve time-varying ICA problem. The results of simulation experiments demonstrate that it’s efficient to apply EASI into invariant mixing, break mixing and time-varying mixing for multiple channel data. Therefore EEMD based single channel EASI (EEMD-EASI) for single channel data is proposed and applied into Wuqiangxi Dam displacement monitoring data for deformation analysis. The result infers that EEMD-EASI could separate displacement component caused by temperature and age factors more accurately. The mixed time-varying matrix presents that the dam became steady several years after starting to store water.