Significance of statistical testing and data integration in acoustic emission analysis

In the applications of structural health monitoring, Acoustic Emission (AE) data can be considered Big Data simply due to the large number of signals acquired or the stochastic relationship between their different variables. However, statistical tests or significance tests are not used in the AE data analysis, neither in the AE signal processing nor in parameter analysis. This study addresses the importance of the significance tests in the AE data analysis (both signal-based and parameter-based approaches). Artificial stress waves are simulated on a stainless steel 304 plate using the Hsu-Nielsen source and Low-Velocity Impact (LVI) events to generate data to enrich the dataset. Statistical tests such as χ 2 and Kruskal-Wallis (KW) tests are used to analyse the signal-based and parameter-based AE data. Statistical analysis minimised the effect of the noise floor in the time-frequency analysis of the AE signals and reduced the sensor effects in the extracted AE descriptors. In addition, the significance tests also revealed the most appropriate method for integrating the AE data acquired from different sensors. • χ 2 significance tests are applied to CWT coefficients. • Significance levels are established to reduce noise floor in CWT analysis. • Kruskal-Wallis tests are used for optimising AE descriptors. • AE data are normalised to reduce sensor effects.