Time domain feature analysis for gas pipeline fault detection using LSTM

This paper presents a novel approach for gas pipeline fault detection using time domain features extracted from acoustic emission (AE) signals. The method leverages basic time domain statistical features including mean, variance, root mean square (RMS), and peak values from AE signals to characterize different pipeline conditions. These features are then processed through a Long Short-Term Memory (LSTM) network to capture temporal patterns critical for accurate fault classification. We evaluate our methodology on the GPLA-12 dataset containing 12 different pipeline conditions and compare the LSTM performance against traditional machine learning approaches such as Random Forest. Results demonstrate that our LSTM model achieves superior classification accuracy while effectively handling the temporal dependencies in acoustic signals. The proposed approach offers a computationally efficient solution for real-time pipeline monitoring systems, as it eliminates the need for complex signal transformations while maintaining high detection accuracy. This research contributes to enhancing pipeline safety monitoring systems by providing a reliable method for early fault detection using simple yet effective time domain features.