Purpose To address the challenge of detecting ultra-small leaks with weak signals in steel–plastic transition joints of polyethylene (PE) pipelines, this study proposes an intelligent classification method for natural gas micro-leaks based on acoustic emission (AE) multi-frequency domain features and random forest (RF) algorithms, enabling effective detection of such micro-leaks. Design/methodology/approach AE signals from pinhole leaks (aperture sizes: 0.1, 0.2 and 0.35 mm) under actual operating pressures (0.1, 0.2 and 0.3 MPa) were collected via AE sensors. After filtering and pre-processing the data set, it was divided into training and test sets in a 7:3 ratio. Features were then extracted from the time-domain, frequency-domain and time–frequency domain. A parameter grid search method was used to optimise the key parameters of the RF model, identifying the optimal combination. Findings The results show that the RF model can not only respond quickly and predict the leakage degree accurately but also the performance evaluation metrics (accuracy, precision, recall, and F1 score) all exceeded 90%, which can effectively identify the weak leakage signal and accurately distinguish the micro-leakage state under different working conditions. Originality/value The research proposes an intelligent classification method for micro-leakage of PE pipeline steel–plastic joints based on AE multi-domain features and RF. The core objective is to accurately and rapidly identify micro-leakage conditions of PE pipeline steel–plastic transition joints under various operating conditions, thereby achieving early warning for micro-leakages in natural gas pipelines.
Qiao et al. (Tue,) studied this question.