Regularization for pipeline impulse response extraction with least square deconvolution
Xun Wang, Mohamed Salah Ghidaoui, Pedro Jose Lee · HAL (Le Centre pour la Communication Scientifique Directe) · 2018
Impulse response function (IRF) is frequently used for pipeline defect (e.g., leakage and blockage) detection. IRF can be extracted via a least square (LS) deconvolution method if the input signal is known. However, the LS deconvolution is usually an ill-posed inverse problem, thus the regularization approach is needed to solve this problem. The determination of the regularization parameter is essential for regularization problems and is strongly problem-dependent (no universal method that always produces robust and good results). In the present paper, prevailing regularization methods Generalized Cross Validation (GCV) and L-curve are used and tested to decide the optimal regularization parameter. Pseudo random binary sequence (PRBS) signal, which switches between two constant levels with noise-level amplitude, is used as input signal. Numerical and experimental examples show the importance of the determination of regularization parameter and justify that the proposed methods are efficient for extracting the IRF with PRBS signal.