Duct acoustic mode identification is easily biased by faulty microphones, whose abnormal channel deviations degrade the performance of traditional discrete Fourier transform and compressivesensing methods. To address this issue, this paper proposes a row-sparse Bayesian integrated detection (RBID) method for simultaneous faulty-microphone localization and duct mode identification. In the proposed framework, abnormal channel deviations and mode coefficients are jointly modeled under a multi-snapshot row-sparse hierarchical prior, and their posterior estimates are inferred within a Bayesian framework. Faulty microphones are then automatically localized from the estimated abnormal deviations, after which a second-stage Bayesian compressive-sensing identification is performed using the remaining healthy channels. Numerical simulations show that, when the proportion of faulty microphones does not exceed 10%, the proposed method achieves fault-detection accuracies above 90% and dominant-mode amplitude errors below 1 dB. Experiments on a 1.5-stage axial compressor further show that, after excluding the detected faulty channels, the dominant-mode amplitude error is around 0.5 dB. The proposed method is effective for duct mode identification under a relatively small proportion of persistent or quasi-persistent faulty microphones, while reducing the need for case-by-case tuning of regularization coefficients.
Wang et al. (Sat,) studied this question.