ABSTRACT Accurate acoustic localisation of surface discharge on converter valve damping capacitors is essential for the safe operation of high‐voltage direct current (HVDC) systems. However, the stochastic nature of discharge signals, leading to frequency ambiguity, coupled with the shielding effects of near‐field obstacles, poses significant challenges to the accuracy and robustness of traditional sound source localisation algorithms. To address these issues, this study proposes a fuzzy narrowband focusing beamforming method for acoustic source localisation. Initially, the time–frequency characteristics of discharge acoustic signals are thoroughly analysed using continuous wavelet transform (CWT). Subsequently, a generalized bell‐shaped fuzzy function is introduced to focus and fuse the fuzzy narrowband, effectively mitigating problems associated with frequency drift and uneven bandwidth. Furthermore, the sound source localisation task is reformulated as a sparse signal recovery problem, and a compressed beamforming algorithm based on Block Sparse Bayesian Learning (BSBL), combined with the Expectation Maximisation (EM) method, is employed to achieve accurate estimation of the direction of arrival (DOA). Finally, the superiority of the proposed algorithm is validated through simulation experiments and simulated valve hall tests. The results demonstrate that the method achieves excellent localisation accuracy and robustness across varying signal‐to‐noise ratios (SNRs), different characteristic frequencies, and diverse microphone array configurations. In field experiments within a simulated valve hall, the proposed method achieves an average spatial angular error of 0.75 with a standard deviation of 0.92, which is significantly lower than that of conventional algorithms. Overall, this research provides an innovative technical approach for the early detection and high‐precision localisation of concealed defects in converter valves.
Yang et al. (Thu,) studied this question.