Research On The Identification Technology For Hypervelocity Impact Iduced Damage Using Sparse Sensor Network

Increasing numbers of space debris seriously threaten the safety of in-orbit operating spacecraft.Large-scale and long-term spacecraft,represented by the Space Sation,is more vulnerable to space debris,thus it is essential to establish an in-orbit perception system to real-time monitor the hypervelocity impact.At present,two modules of in-orbit perception system,i.e.,impact detection and localization,have been practically used in engineering,while the damage recognition module has not been applied yet.In addition to the difficulty of accurately extracting signal characteristics and low algorithm reliability,the small damage recognition range is also one of the important reasons.Existing research on damage recognition mostly focus on the plate with the size smaller than 600 mm,far less than the layout distance of the actual sensor network.Based on which,hypervelocity impact tests were conducted to launch Φ3.2mm aluminum alloy spherical projectiles to impact aluminum alloy plates and reinforced plates with the size of 2000×500×2.5mm.This paper studied the generation and processing method of acoustic emission signal,selection of damage chara cteristic parameters,and damage recognition approach using sparse sensor network.On this basis,a new method was proposed based on convolutional neural network for the identification of damage at long distance.Firstly,a variety of feature parameters were extracted to enhance damage recognition using sparse sensor network.The acoustic emission signals were obtained through numerical simulation and hypervelocity impact experiments and processed with modal acoustic emission theory,wavelet decomposition and wavelet packet decomposition.From the perspective of the amplitude and energy,the paper selected the characteristic parameters of the target plate damage,such as S0 mode peak and valley value,the frequency band energy proportion of wavelet packet decomposition,etc.A variety of feature parameters were combined to enhance the accuracy of feature parameters for damage pattern classification at long distance.Secondly,a convolutional neural network was established to realize impact damage recognition using sparse sensor network.Based on the numerical simulation signals,identifying the cratering-or-perforation,crater depth and hole diameter within the monitoring distance from 150 mm to 1710 mm.The results showed that the recognition accuracy of three damage types was respectively 99%,97% and 80% for aluminum alloy plates and 97%,92% and 75% for aluminum alloy reinforced plates.Finally,based on the data obtained from the hypervelocity impact experiments,the damage pattern recognition capability of proposed convolutional neural network at different distance was verified.For the aluminum alloy plate and reinforced plate,the convolutional neural network was trained and verified 3 times.The results showed that the recognition accuracy can reach above 86.67% for both plates,and the error samples were distributed on the speed corresponding to the perforation damage,which showed that the recognition accuracy of the convolutional network for the cratering-or-perforation mode was higher than that of the perforation diameter,which corresponded to the results in the simulation.The error samples of the plate were mostly distributed at long distance,and the error samples of the reinforced plate were distributed at all distance.The results have significant reference for the damage recognition of space debris induced damage on the surface of large spacecraft.