Open Access Open Access  Restricted Access Subscription or Fee Access

Application of Shape-based Similarity Measures to Classification of Acoustic Emission Waveforms



Over the past decades, Acoustic Emission (AE) has emerged as a promising in-situ inspection technique in different applications such as civil or airborne structures, rotating machinery, or process monitoring. As AE refers to the sudden release of elastic energy due to structural changes in stressed materials (i.e. damage), the acquired waveforms are related to their underlying source mechanism. Thus, using suitable signal processing techniques, different source mechanisms can be distinguished. Despite encouraging results, processing of AE signals still remains challenging due to high sample rates. Therefore, in practical applications typically only few parameters (i.e. hit rate, rise time, maximum amplitude) are extracted using threshold-based approaches. Thus, storage requirements are greatly reduced. However, to develop an indepth understanding of AE source mechanisms, detailed waveform analysis is generally required. Here, frequency domain approaches are extensively used. While competing methods use different statistics, cross-correlation, or autoregressive coefficients, detailed investigation of raw AE waveforms in time domain has not received much attention. In time series data mining, shape-based distance measures are frequently used to assess the similarity of time series objects. Considering each time series as a high dimensional feature vector, the distance of two time series can be computed using Lp-norms such as Euclidean distance. This approach is highly attractive, as it is parameter- free. Due to strict point-to-point comparison these measures are sensitive to scaling and time warping (i.e. non-linear shifting on the time axis causing local stretching or compression of the signal). While the former can be addressed by suitable normalization of the data, Dynamic Time Warping (DTW) is frequently used to achieve invariance of distance calculations to warping. Thus, DTW allows compensating for misalignment of individual time series elements. In this work, experiments are performed using thin plates of composite material. Waveforms related to AE hits from different underlying source mechanisms (i.e. in-plane and out-of-plane source motion) are acquired. Results of hierarchical clustering and nearest neighbor-based classification indicate that shape-based distance measures are suitable to distinguish between different AE sources using raw AE waveform data.


Full Text:



  • There are currently no refbacks.