Principal Component Analysis in Multivariate Microphones Response to Simulated Leakage in Metal Pipeline of Compressed Air

Luiza B. Fernandes, Rejane B. Santos, and Ana M. F. Fileti


Signal Processing, FFT, leakage, PCA


Systems capable of diagnosing and characterizing a leakage in pipes are important not only to avoid losing material, but also for human health. For instance, in a building, a leakage of inflammable gas through pipes may lead to huge fire injuries or intoxication. For analysing the existence of gas material escaping from pipes, the acoustic method acts investigating the sounds signals in situations without leakage and in situations with simulated leakage. Microphones coupled to the lab scaled gas distributor pipe capture the acoustic signals in time which are further decomposed in their frequencies components by Fast Fourier Transform (FFT). Thus, a principal component analysis (PCA) is applied considering the amplitudes, in the respective frequency accused by the FFT method, as variables. It is revealed that it is possible to represent the system in smaller dimension size in which much of the original information preserves. For instance, plotting data in three dimensions, in terms of the three principal components, the representativity, or explained variance, is greater than 70% and some patterns become visually enlightened. In fact, 78 components are enough to get representativity greater than 90% against the original data that had the observations distributed in 6599 different amplitudes in frequency domain. In time domain, it is possible to note that the average absolute amplitude response from the microphones increases after disturbance but the farther it is its position related to the source of disturbance, the less it tends to increase.

Important Links:

Go Back