The aim of the study is to develop a method for controlling the sending of an artillery shell to the gun’s chamber using an acoustic portrait. The existing method for controlling the delivery of artillery ammunition to the gun’s chamber with a separate loading method is based on measur-ing the speed of one of the elements of the rammer. Such an approach to the control didn’t provideguaranteed reliability due to the impossibility of measuring speed in the final segment of the pro-jectile’s inertia motion. At present, vibroacoustic methods of analysis are widely used in various fields of science and technology and can be extended to the problem under consideration. The essence of the method proposed in the article is to excite acoustic vibrations in the "projectile - gun chamber" system and to distinguish characteristic acoustic portraits (signatures) with their subsequent analysis. To study this method, a laboratory bench has been developed that imitates the barrel of a gun with a chamber, and a shell simulator with various obturator belts. A projectile impact at the moment of jamming in the chamber cone or applied externally, for example, on the gun’s body, excites characteristic acoustic vibrations, which differ for cases of reliable and insuf-ficient sending. In the developed stand, acoustic vibrations were excited by an external impact on the resonator and were recorded for subsequent analysis. For an unambiguous classification of events of reliable jamming and insufficient submission, it is necessary to select the optimal vector of signs of an acoustic portrait of the obtained audio recordings. The usual spectral conversion makes it possible to distinguish characteristic frequencies, however, the set of such spectral com-ponents is not suitable as classification features due to the significant array of data obtained as a result of this analysis, and also due to the inability of the Fourier transform to recognize short-term low-power bursts. Therefore, as the classification features were selected mel-frequency cepstral coefficients. Based on the set of such coefficients, using the artificial neural network, the degree of jamming of the projectile simulator in the stand was classified into three categories: “sleep-row is not jammed”, “insufficient shell projectile”, “projectile jammed”. As a result of training the neural network on a significant sample of audio recordings, a classification accuracy of 90% was achieved. It is shown that the developed method of such vibroacoustic analysis can be applied in robotic artillery weapon control systems, as well as in other technical tasks, for exam-ple, in oil and gas production to control the docking of articulated main pipes.
Read full abstract