The Method Of Pattern Identification Based On AE For Damage Of The Aluminum Alloy Plate Caused By Hypervelocity Impact

As more and more space activities being taken place, the space debrisenvironment is getting worse, which has constituted a serious threat to the safety ofthe spacecraft on orbit. In order to real-time monitor the spacecraft impacted by thespace debris, it is of great significance to establish the onboard monitoring system.The ability to identify the damaged pattern of the spacecraft structure caused byhypervelocity impact is an important function of the onboard monitoring system. Theonboard monitoring system which based on the acoustic emission technology iscurrently considered to be one of the best methods, yet the study on this method foridentifying the damaged pattern of the spacecraft structure is not enough so far.Based on the background mentioned above, the paper took aluminum alloy plateswhich were normally hypervelocity impacted by aluminum alloy spherical projectileas research objects; and set a serious study, which includes the methods to acquire andanalyze the acoustic emission signals, measurements of the signals’ characters,collection and selection of the characteristic parameters of acoustic emission signal aswell as methods to identify damaged patterns. At last a method which based on thenearest neighbor classification or neural networks, to identify the damaged pattern ofaluminum alloy plate caused by hypervelocity impact, was proposed.The research range of this paper:3mm thick aluminum alloy plates werenormally impacted by aluminum alloy spherical projectiles which diameter between1~5mm, with speeds between100m/s to5000m/s.Firstly, hypervelocity impact acoustic emission signals were processed withwavelet analysis and the concept of oscillation energy. The study showed that theoscillation energy of acoustic emission signals and wavelet decompositions a4(db10)、d4(sym8)、d3(ciof5) in the first45μs, could properly characterize the damage patternof the aluminum alloy plate. Besides comparing to parametric approach and powerspectral analysis, processing hypervelocity impact acoustic emission signals withwavelet analysis and the concept of oscillation energy could get more usefulinformation.Secondly, the author used the method of nearest neighbor classification andneural networks to identify whether aluminum alloy plate has been perforated of not,the diameter of projectiles and perforated holes respectively. The result ofidentification indicated that the method of nearest neighbor classification and neuralnetworks could effectively identify whether aluminum alloy plate has been perforatedor not, and the average accuracy were more than96.00%and93.00%respectively.For the identification of diameter of projectiles using the method of nearest neighborclassification and PNN, when the projectile’s diameters of validation samplesbelonged to that of the training samples, the average accuracy were87.57%~96.32%and96.27%respectively. When they weren’t, the average accuracy were81.73%~83.06%and78.44%respectively. For the identification of diameter of perforated holes using the method of nearestneighbor classification and neural networks, when the projectile’s diameters ofvalidation samples belonged to that of the training samples, the average accuracywere96.24%~96.69%and89.77%~93.36%respectively. When they weren’t, theaverage accuracy were89.77%~93.36%and81.06%~82.75%respectively.In the end, through comparing the average and highest accuracy of the method ofthe nearest neighbor classification and that of the neural networks, the differentrecognition performance of the two methods was presented. And the advantages anddisadvantages of the two methods applied in the field of the paper were analyzed. Theresult indicated that both the method of nearest neighbor classification and neuralnetworks could effectively identify whether aluminum alloy plate has been perforatedor not. For the problem of identifying diameter of projectiles and perforated holes, theidentifying accuracy of the nearest neighbor classification method were better thanthat of the neural networks method, and the recognition performance of method ofdistance-weighted nearest neighbor classification was best among patternrecognition method discussed in this paper. Compared to the methods of nearestneighbor classification and the probabilistic neural networks (PNN), the identificationresults of BP neural networks have the feature of randomness, and this feature limitsBP neural networks’ application in the field of damaged pattern identification ofaluminum alloy plate caused by hypervelocity impact.The result of this paper has reference signification for developing the technique ofidentifying the damaged pattern of spacecraft complicated structure caused byhypervelocity impact; it also provides technical supports to the onboard monitoringsystem aimed at the space debris. It is valuable to engineering application andtheoretical guidance.