Abstract Pipeline leakage detection in boiler energy systems is essential for operational safety and efficiency, yet conventional techniques such as pressure-based and mass-balance methods often lack real-time performance and sensitivity to small leaks. Although acoustic emission (AE) technology offers dynamic, non-destructive monitoring, its practical application is hindered by noise interference and limited training samples under industrial conditions. This paper introduces an enhanced support vector machine (SVM) framework designed for robust AE-based leakage detection. The proposed approach integrates three key contributions: first, a multi-domain feature fusion strategy that combines time-domain and frequency-domain parameters for enhanced signal separability; second, a spectral sparsity-guided dynamic kernel selection mechanism that adaptively optimizes the model for varying signal characteristics; and third, a margin-based boundary sample weighting strategy that mitigates the influence of noise near the hyperplane. Experiments involving three leakage types—spot, fracture, and explosion tube—were conducted under both low-noise (40 dB) and high-noise (70 dB) conditions. The model achieved perfect classification (100% accuracy) under quiet settings, and maintained accuracies of 92.3%, 88.1%, and 85.4% for the respective leak types under noisy conditions, outperforming conventional SVM by 12–15%. These results demonstrate that the proposed framework significantly improves detection reliability in noisy, data-scarce environments, providing a practical tool for early leakage identification in industrial boiler systems. Future work will focus on adaptive noise modeling and online threshold learning to further enhance the framework’s robustness and adaptability in dynamic industrial settings.
yuan et al. (Sat,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: