Acoustic feature processing strategy for leak degree identification in non-metallic pipelines

In response to the challenge of accurately identifying the degree of leaks in non-metallic pipelines, this study proposes acoustic feature processing strategy based on the Principal Component Analysis (PCA) and Bidirectional Long Short-Term Memory Networks (Bi-LSTM). The hydrophone sensors are deployed to collect acoustic pressure signals generated by different degrees of pipeline leak. Acoustic signal features of varying leak degrees are extracted from the perspectives of the time domain, the spectral domain, and waveform shape. Subsequently, PCA is employed to streamline and reduce the dimensionality of the extracted features. The optimal number of principal components is ascertained based on the Cumulative Variance Contribution Rate (CVCR), which eliminates redundant features and preserves principal components. Furthermore, the PCA-Bi-LSTM model is designed to identify the degree of pipeline leak. The principal components are organized chronologically into cells for input into the Bi-LSTM. Comparative analysis is performed using different statistical features and neural network architectures, with additional experiments conducted under varying pressures and leak locations to validate the generalizability of the proposed method. The results indicate that the PCA-Bi-LSTM approach achieves a leak identification accuracy of 98.6 %, which is significantly higher than the 91. 5 % accuracy of the unidirectional LSTM network and the 93.2 % accuracy of the traditional Bi-LSTM network. The results demonstrate that the designed strategy effectively improves the accuracy of identifying the degree of non-metallic pipeline leaks and demonstrates a certain level of generalizability.