Reducing sensor effects in Acoustic Emission signals via a novel hybrid two-block blind source separation framework

This study presents a novel, two-block hybrid framework aimed at reducing sensor-induced artefacts and distortions in single-channel Acoustic Emission (AE) or stress wave signals. The novelty of the work lies in the hybridisation of a machine learning framework and a cepstral analysis framework to process single-channel stress wave signals. The first block employs Independent Component Analysis (ICA) to separate the statistically independent Time Frequency (TF) features of the source signal from sensor-induced distortions. This is achieved by grouping the independent components corresponding to the source signal using negentropy. The second block employs the Cepstral deconvolution process. Here, the signal is converted into the cepstral domain, and the low- and high-quefrency components corresponding to reverberations and sensor-induced bias are suppressed. The proposed framework is tested against stress wave signals propagated in thin stainless steel plate and acquired by four sensors that differ in sensitivity and operating frequency. The recovery of the source signal is validated by comparing the experimental stress wave signals with the theoretically modelled Lamb waves. The spectrogram-based results demonstrate that the proposed framework enhances the recovery of the source signal. Quantitative evaluations are also conducted by measuring permutation entropy, Time of Arrival (ToA), and power of the signals. The findings confirm that the framework improves the reproducibility and the interpretability of the signal while preserving its transient characteristics. In particular, the shorter ToA is an indicative of an accurate identification of the signal arrival, whilst the increased entropy is an indicative of the dynamic behaviour of the source signal. The outcomes of this research indicate that the proposed two-block framework constitutes robust signal preprocessing, which is essential for improving the interpretability of AE testing.