Research Of Spacecraft Damage Assessment Method Based On Machine Learning

With human's exploration of space,there are more and more debris in the space and these debris can affect the normal operation of spacecraft.Hence,it is necessary to assess the damage of hypervelocity impact.There are many methods for damage assessment of spacecraft at home and abroad,and Infrared thermal imaging technology is applied to realize damage assessment in this paper.According to characteristics and requirements of hypervelocity damage assessment,some algorithms are designed based on infrared image sequence of aerospace composites,the main research contents of this paper are as follows:1.Since the difference of thermal medium in different damaged areas,the temperature change in different damage areas will be different during the heating process,and this can be used to judge different damaged areas.To start with,some physical attributes are proposed to describe thermal response curves to mine the physical information.After that,Na?ve Bayesian classifier and Weighted Bayesian classifier are built to distinguish different thermal response curves.In addition,the environment of damage assessment may be variable.Hence,a multiple environment Bayesian classifier is proposed to assess the damage in complex environments.2.The above method can judge damaged areas by thermal response curves.In order to realize the damage visualization in the assessment process,a damage assessment framework based on image is proposed.According to the infrared image sequence of hypervelocity impact,we improve the traditional linear model to get a damage reconstruction model,and the model is solved by the variational Bayesian to obtain reconstructed damage images.After that,active contour model is used to extract more accurate damaged areas.Moreover,in order to extract the damage area more accurately,we improve the active contour model based on the characteristics of infrared image of hypervelocity impact from two aspects: considering the temperature change rate of different areas(damaged and non-damaged areas)and considering the interaction of different segmentation factors(local and global).In addition,in order to obtain the location of damage on subsurface,we use multiscale transformation to fuse damage in different layers.3.The above methods are used to determine whether there is damaged areas and the location of damaged areas.In order to get more objective and accurate damage description,physical information and morphological information are proposed to quantify the damage.By establishing quantitative attributes,we can describe each damage digitally.Moreover,in order to judge different types of damage,ensemble learning classifier is established based on these quantitative attributes.