Machine Learning in Predicting the Performance of Acoustic Emission Signal Filtering Methods: Association Between Predictors

This article presents the results of assessing the association between the quantitative values of the acoustic emission signal processing system predictors obtained at the output of digital filtering methods that were approximated based on the second-degree Butterworth and Bessel polynomials, as well as the Daubechies wavelet function of the eighth smoothness level. The measured values of the data processing system predictors based on the machine learning method - linear regression, were subjected to assessing their association. The association between the system predictors capable of determining, identifying and characterizing at the predictive analytics level the method that effectively processes noisy signals during filtering was also assessed. The regression analysis method revealed an association between the data processing system predictors associated with a simultaneous improvement in the system properties in terms of noise immunity and accuracy of acoustic emission signal processing in comparison with the Bessel filters and the Daubechies wavelet of the order in question. Moreover, the effectiveness of machine learning in assessing the relationship between predictors at a quantitative level is confirmed by calculating the correlation and determination coefficients and testing statistical significance.