Blind identification of pipeline damage using independent component analysis with wavelet transform

摘要

This study discusses output data identification algorithms for pipeline structural defects using independent component analysis; a powerful tool for solving blind source separation (BSS) problem. Sparse like features are discovered in the three-axis magnetic field data of a pipeline use to be the hidden targeted independent sources that indicate damage information of the structure under observation. Wavelet transform algorithms are applied on the 3-axis magnetic field data. The output signals are cast into the blind source separation model where FastICA algorithms are applied on the wavelet-domain mixtures to separate them into their respective independent components. Sharp spikes are found in these independent components that clearly show the time instant of the damage occurrence. The location of the pipeline damage can be found by exploiting the (time-based) temporal information contained in the spatial signature of the recovered mixing matrix. WT-ICA method has been applied on synthetic data of a twelve degree of freedom time-varying system where damage is modeled by abrupt stiffness variation. Laboratory experiments and real world underground pipeline data were recorded and fed as mixtures in to the WT-ICA based BSS model. Results provide clear physical interpretation of the pipeline structural damages subjected to various kinds of stresses.