Identifying acoustic signature of inflow control valve’s condition Using distributed acoustic sensors

In this paper, we present a novel method to identify the acoustic signature of Inflow Control Valve’s conditions and classify them. The proposed method consists of three stages: preprocessing sounds data, acoustic feature extraction and multi-class classification. In the preprocessing stage, we applied power normalisation to smooth the acoustic signals and then fed the normalized acoustic data into feature extraction algorithms. We analysed the series of acoustic features in time domain, frequency domain and also in an unsupervised feature extraction algorithm. In time domain, we performed an extensive feature statistic analysis by comparing six audio features and selected the best one. In frequency domain, the features from wavelet transform was extracted. In addition, acoustic data is converted to frequency domain by applying short time Fourier transform and its output fed into Principal Component Analysis algorithm. Our proposed method combined all extracted features from different methods and composed the novel feature set. In the last, two classification algorithms, Artificial Neural Network and support Vector Machine, are implemented to test and validate the novel feature set. We evaluated our method by performing an experiment on seven real word datasets and experimental results demonstrated its superior performance compared to other method.