Improving Ultrasonic Testing by Using Machine Learning Framework Based on Model Interpretation Strategy

Chinese Journal of Mechanical Engineering

Table 1 Indexes and implications of 20 important features

Feature index	Feature name	Feature index	Feature name
P₁	Maximum amplitude of time-domain signal	P₁₁	The low frequency corresponding to the − 6 dB bandwidth of frequency spectrum
P₂	ToA of maximum amplitude	P₁₂	The high frequency corresponding to the − 6 dB bandwidth of frequency spectrum
P₃	Temporal kurtosis	P₁₃	Resonant frequency
P₄	Minimum amplitude of time-domain signal	P₁₄	Spectral mean
P₅	ToA of minimum amplitude	P₁₅	Spectral variance
P₆	Temporal mean	P₁₆	Spectral skewness
P₇	Crest factor	P₁₇	Spectral kurtosis
P₈	Temporal margin index	P₁₈	Energy ratio of 1st wavelet sub-band
P₉	Spectral peak value	P₁₉	Energy ratio of 2nd wavelet sub-band
P₁₀	Center frequency	P₂₀	Energy ratio of 3rd wavelet sub-band