Full Waveform Inversion Of Asteroid Internal Structure Based On Radar Data

With the development of deep space exploration,asteroids have become one of the main targets of deep space exploration in the 21st century,and the tomography of asteroids and the composition analysis of asteoid are the main important tasks for asteroid detection.Revealing the internal structure,material composition and other characteristics of various asteroids is of great scientific significance for many aspects,such as:studying the origin and evolution of asteroids,searching for outer space minerals and water,and defening the earth from asteroid impacting.At present,China is actively preparing a deep space exploration project for asteroids,and plans to launch the first probe“Zheng He”in 2024,targeting the near-Earth asteroid 2016HO3 with a diameter of 40?100m.Although since the 1990s,space agencies of the United States,the European Union,and Japan have launched a series of probes to conduct close-range asteroid detection activities through flybys,orbits,landing sampling,and impacts,the asteroid tomography has never been carried out in the past.Therefore,asteroid tomography is a new field,and the research related to the theoretical foundations and detection methods needs to be done in advance.Considering the limited observation conditions of asteroid missions,the observed density may not be guaranteed.We propose to use the electromagnetic full waveform inversion(FWI)which can achieve high resolution imaging under the condition of relative low observation density to do the asteroid tomograohy.The main difficulties in the practical application of FWI are as follows:1)Due to the expensive computation,FWI is usually performed by using the gradient-based local optimization method,which makes that the FWI is easy to fall into local minima.This shortcoming is particularly obvious when the bandwidth of the transmitted signal is limitted.2)Full waveform inverion needs to simulate the propagation of wave field and store the wavefield data,which demands huge amount of computation and computing resources in 3D inversion.In addition to the technical difficulties mentioned above for FWI itself,there are also some particular technical difficulties for asteroid exploration missions,such as:1)The observation density and positioning accuracy of measurement points may not be guaranteed due to communication delay,energy limit,weight limit of probe and other factors.2)The material composition of the asteroid could be dispersive and anisotropic,thus the conventional FWI may not be able to accurately image the inner structure of the asteroid.3)Cosmic microwave background radiation and solar electromagnetic radiation can reduce the signal-to-noise ratio of radar signal.4)prior knowledge of the inner structure of the asteroid is very limitted,so we cannot use prior knowledge to constrain FWI and mitigate the multisolvability.In view of the aforementioned difficulties in the application of FWI to asteroid tomography,the contents of this thesis mainly consist of the following aspects:(1)Research on robust FWI methods of radar data for asteroid tomography grahy.On the basis of the traditional electromagnetic FWI algorithm,the envelope inversion is added to provide a more accurate initial model for the FWI of the asteroid's internal structure.The W₂metric(quadratic Wasserstein metric)in the optimal transport theory is used to construct an objective function with stronger“convexity”.Numerical experiments show that,the above two methods can effectively improve the robustness of imaging of the internal structure of asteroids without sacrificing resolution.(2)Research on FWI algorithm of dispersive medium asteroids.According to existing studies,many common rocks(such as granite,basalt,pyroxene and plagioclase,etc.)have certain dispersion characteristics in the radar band(MHz?GHz),And their dispersion characteristics can be described using the Debye dispersion model.Considering that asteroids may also have dispersive properties,the FWI method for Debye dispersive media is developed in this thesis.Numerical experiments show that FWI can accurately reconstruct distribution of the four electrical parameters in Debye dispersive medium,which verifies the feasibility of FWI for imaging the internal structure of asteroids with dispersion characteristics.(3)Analysis and optimization of observation system and radar parameters.According to the asteroid exploration missions,the“orbiter to lander”bistatic radar system is considered to be one of the most feasible observation modes for asteroid tomography.Based on the“orbiter to lander”observation mode,this thesis analyzes the influence of measurement point density,distribution of orbit,signal-to-noise ratio,radar bandwidth and carrier frequency on the asteroid tomography results.Three-dimensional numerical experiments show that,in the case of high signal-to-noise ratio,even under sparse observation(such as only about 20 measurement points around the asteroid)or the measurement orbits cover only apart of the asteroid(such as the covered latitude range is only 10°),FWI can still reconstruct the internal structure of asteroids with high accuracy;But if the signal is heavily contaminated by noise,intensive observations are required in order to ensure the accuracy of the inversion.(4)High performance computation for imaging the asteroid's inner structure.Since the FWI of 3D asteroid requires a huge amount of computation,this thesis,based on the electromagnetic finite difference time domain(FDTD)method,uses CUDA parallel architecture and MPI communication standard to achieve high performance parallel computation of electromagnetic FWI with multi-node and multi-GPU.Numerical experiments show that the computing efficiency of multiple GPUs is related to the model size.For the small model,due to the limitation of communication bandwidth between nodes,the data transmission time between nodes is longer than the wavefield calculation time,thus multi-GPU will not significantly improve the computing performance comparing to single GPU.For the big model,the data transmission time between nodes is less than the wavefield calculation time,thus the data transmission can be better hidden in the computation,and the performance improvement of multi-GPU is more significant.