Research on Weak Vibration Characteristics Based on EMD and Design of MZI Classifier
Hao Feng, Ming Wang, Sha Zhou, Lipu Du, Dunzhe Qi
Abstract
Weak vibration identification is one of the key challenges in the field of pipeline safety warning using distributed fiber optic sensor. In this paper, we proposed a novel algorithm that experimentally exhibits better performance concerning weak signal recognition over conventional programs. Empirical Mode Decomposition (EMD) was employed in the method considering that the temporal and spectral attributes could be better preserved and magnified in an intrinsic form than the original signal. After extensive observation and analysis, the second and third Intrinsic Mode Function (IMF) exhibited better consistency and identifiability among samples belonging to the same and different categories. As a substitute for the original signal, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">IMF</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">IMF</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sub> were jointly utilized as the research target, where a series of carefully selected parameters were applied to evaluate the feasibility of the IMF features for weak vibration identification, giving birth to our classification vector <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">V</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> . Combining <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">V</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> with certain machine learning algorithms, precise identification of noise and five typical pipeline vibration signals was achieved. A conventional vector <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">V</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> and a deep learning network were also provided as the control groups for two mainstream algorithms. Experimental results show that the val_accuracy of SVM- <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">V</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> is 7.77% higher than <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">V</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> . In terms of precision and recall, SVM- <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">V</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> exceeds <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">V</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> by 10.83% and 7.97%, and surpasses deep learning by 8.58% and 7.97%.