Research On Explosion Shock Wave Data Processing

Overpressure test is one of the key indicators to assess the power of the weapon explosion damage.How to reliably and effectively collect the overpressure value and perform certain signal processing on the captured data to improve the accuracy of the test data is the key to shock wave testing,and it is also a prerequisite for related follow-up research.According to the storage test method,this paper designs an explosion shock wave overpressure storage test system,completes a series of static explosion experiments,and captures a large number of ground measurement points pressure data under different proportional distances.In view of several practical problems encountered in data processing,this paper carries out research in three related directions.The first problem is that the captured data aliased a large amount of environmental noise.We use the wavelet method to analyze the frequency bands of the signal,establish a noisecontaining shock wave model based on the analysis results and empirical formulas.compared the performance of empirical mode decomposition(EMD)and wavelet threshold denoising method in the model with noise by experiments.The results show that the EMD denoising algorithm can achieve a better denoising effect,and because the algorithm has self-adaptability,the influence of human factors on the denoising effect is reduced,and the processing efficiency is significantly accelerated.Finally,the actual captured signal is de-noised by EMD,and the application effect of this method in practical engineering is verified.The second problem is that gross error discrimination method based on statistics has rigid requirements for the minimum sample size.At the same time,such methods may have misjudgments when dealing with overpressure peaks.this paper proposes a similarity-based discrimination method.This method uses the normalized similarity as the basis for whether the overpressure peak value of the captured data is a gross error value.it can process a single data,and provides a new idea for judging the gross error of shock wave overpressure value.We verified the feasibility of two calculation methods of DTW similarity and Euclidean distance similarity in the experiments of this paper.The third problem is that the captured pressure signal has outliers,platforms,serious distortion,etc.This paper introduces a machine learning classification algorithm to identify pressure signals.Through the analysis of the measured data,the captured waveforms are divided into three categories,which are normal shock wave signals,presence platform signals,and attenuated and severely distorted signals.We extracted 10 types of data features such as positive pressure time,specific impulse,and DTW similarity from the captured waveform.After that,we trained four classification models: SVM,decision tree,Ada Boost decision tree,and random forest.By comparing and analyzing the evaluation indicators of the models after tuning,the random forest performed best in terms of prediction accuracy,stability,and calculation efficiency.Experimental results prove that the model can effectively and reliably realize the recognition and classification of shock wave signals,and Meet the needs of automatic and rapid signal identification in actual work.