| The explosion shock wave test is a dynamic test,which can evaluate the damage effectiveness of weapon.It is very important to test the parameters of shock wave accurately for weapon development.Shock wave signal is a kind of non-stationary random transient signal,which has the characteristics of fast propagation speed,wide spectrum coverage and wide amplitude dynamic range.Therefore,higher requirements are put forward for the dynamic performance of the shock wave test system.Due to the limitation of manufacturing technology,the working bandwidth of pressure sensor becomes the bottleneck of shock wave testing system.In order to improve the dynamic performance of the pressure sensor and reduce the dynamic error of the test system,this paper introduces Generative Adversarial Networks(GAN)to build the dynamic compensation model of the sensor,which mainly solves the problems of insufficient data and insufficient accuracy of the model.Deep learning algorithms need to be a large number of training data as a premise,in shock wave test,the data acquisition process can cause damage to the pressure sensor,only through the test data to build a data set,can cause serious loss to the sensor,will cost a lot of manpower and material resources as well as introducing more unnecessary human error,affect the quality of the data,resulting in data cannot be used,so this paper introduced to Deep Convolutional Generative Adversarial Networks(DCGAN)enhance the test data.Combined with the excess training method and the optimization of the loss function,the diversity and data characteristics of the data are further optimized.The results show that the enhanced data has good diversity and contains the inherent resonant frequency points of the target sensor,which can be used to establish the dynamic compensation data set of the sensor.Compared with the original test data set,the compensation result of the data set constructed by the supplemented enhanced data reduces the overshoot by about 5.28%,and the rise time is improved 16.00μs.Based on the Speech Enhancement Generative Adversarial Network(SEGAN),the generator is improved for dynamic compensation.Finally,the dynamic compensation model of the sensor is obtained by combining the enhanced dynamic compensation data set.The model training process is based on the shock tube data.The test shows that the obtained model can achieve the reduction of the original signal overshoot from about 119.82% to about5.03%,and the rise time is maintained at about 2.00μs.After the muzzle shock wave signal is compensated,the peak overpressure decreases from 0.1119 to 0.0968,which is closer to the actual shock wave.The dynamic compensation results show that this method can effectively suppress the resonant frequency of the sensor,and the characteristic parameters of the shock wave can be obtained from the compensated data,which proves that the dynamic compensation model can effectively improve the dynamic response performance and test accuracy of the pressure sensor.The Generative Adversarial Networks in Deep Learning is applied to the field of dynamic compensation,which proves the feasibility and practicability of data enhancement,and the generalization and superiority of dynamic compensation.The compensation model is easy to adjust and adaptable,which makes the dynamic compensation simple,intelligent and high precision. |
