基于数据驱动多尺度表征爆破信号特征提取研究OA

To eliminate the influence of noise in blasting vibration signals on signal feature extraction,a data-driven blasting signal denoising method is proposed using Variational mode decomposition(VMD)algo-rithm and Cramér-von misse(CVM)statistic.This method uses the CVM metric of statistical distance to select the main noise modes,and locally uses the CVM statistic on the remaining modes to test the relationship be-tween the modes and the estimated noise distribution.Modes closer to the noise distribution are discarded to obtain more realistic signal fluctuation characteristics.Subsequently,refined feature extraction is performed on time-frequency distribution,instantaneous energy,and marginal energy distribution of the denoised signal.The results indicate that the CVM-VMD method has superior mathematical and theoretical frameworks,making it highly robust to noise and mode mixing,and has unique advantages in blasting signals denoising.The time-fre-quency spectrum of blasting signals exhibits significant characteristics of uneven energy distribution,domi-nance of low-frequency components,and attenuation of high-frequency components.The instantaneous energy spectrum can intuitively display the changes in signal energy at different time points,while the marginal ener-gy spectrum can more accurately reflect the fluctuation of signal energy with actual frequency,which provides important data support for studying blasting mechanisms and evaluating the vibration effects on buildings.