Small-Leak Identification Method Based on Infrasound and VMD–HMM for Pipelines

Pipeline leaks cause not only significant wastage of resources but also economic losses and major safety accidents. The detection of pipeline leaks and the identification of their magnitudes have attracted significant research attention. A promising method for leak detection involves the use of infrasound sensors to detect infrasound waves from leaks. However, the detection of leakage infrasound is highly challenging due to strong environmental disturbances, such as external vibrations and noise, which often obscure weak leak signals. To address these challenges, this study proposes an innovative variational mode decomposition-hidden Markov model (VMD-HMM) method for processing infrasound sensor outputs in noisy environments. The method combines an optimized VMD algorithm with the HMM, leveraging the VMD algorithm’s ability to handle non-stationary signals and the HMM’s superior classification performance for time-series data. The optimization of the VMD algorithm significantly improves noise reduction, while the integration with HMM enhances the accuracy of leak identification. Subsequently, a comprehensive pipeline experimental platform was constructed to validate the proposed method under varying leakage calibers and noise conditions. The experimental results demonstrated that the accuracies of traditional methods, including the HMM, BP neural network, and support vector machine (SVM), did not exceed 80%. In contrast, the proposed VMD-HMM method achieved a remarkable accuracy of 94%, significantly outperforming existing approaches and providing a robust solution for identifying small leaks in pipelines.