Professional Documents
Culture Documents
High-Precision Adaptive Predictive Entry Guidance For Vertical Rocket Landing
High-Precision Adaptive Predictive Entry Guidance For Vertical Rocket Landing
real time, decreasing the traditional predictive algorithms’ guidance error from kilometer- to meter-scale precision.
Nomenclature 2010s. After the space shuttle and manned spacecraft, the Falcon 9
CD = drag coefficient rocket was developed by SpaceX; this rocket is the first orbital rocket
CL = lift coefficient that has been reused and offers a new RLV solution. On 23 December
D = drag force, N 2015, the Falcon 9 rocket completed its first recovery mission with a
KD = drag aerodynamic acceleration deviation coefficient vertical landing at the launch site after stage separation. On 8 April
KL = lift aerodynamic acceleration deviation coefficient 2016, in the SES-10 mission, the rocket landed on a ship for the
L = lift force, N first time. On 30 March 2017, SpaceX used the reusable core vehicle
r = radial distance, m (RCV) that was recovered from the CRS-8 mission to complete a
t = current guidance time, s mission, and subsequently recovered it again. In recent years, many
ts = stop guidance time, s companies have unveiled new RLVs, such as New Shepard of
α = angle of attack, rad Blue Origin and the new-generation Long March rocket of China. With
αk = angle-of-attack coefficient the need for future rocket development, the robustness, reliability,
β = sideslip angle, rad and autonomy of entry guidance systems are increasingly gaining
βk = sideslip-angle coefficient attention.
γ = flight-path angle, rad The Falcon 9 rocket exhibits an innovative approach for solving
θ = longitude, rad the RLV controllable recovery problem [1]: the rocket is equipped
ρ = density with a grid wing and an expandable support structure. This extra
σ = bank angle, rad equipment ensures that the RCV is upright during entry into the
ϕ = geocentric latitude, rad aerosphere during the descent phase and landing phase. This means
ψ = heading angle, rad that the control method for the RCV is distinct from the bank-to-turn
(BTT) method for the space shuttle and manned spacecraft in flight.
By deploying four grid wings around the RCV, the rocket can
Subscripts
simultaneously control both pitch and yaw attitude. To solve the
f = value of a variable at termination of entry phase problems of high dynamic pressure and heat during the atmospheric
m = value of a variable in onboard mode entry phase and to make a precise landing, the RCV must burn two or
t = waypoint index three times to slow down. Both returns to the launch site and the ship
are shown in Fig. 1.
As shown in Fig. 1, typical RCV recovery processes can be divided
into two types: recovery at the launch site and recovery on a ship.
I. Introduction Obviously, the method of recovery at the launch site requires burning
guidance method has evolved to exhibit great potential. The most dθ V cos γ sin ψ
notable strengths of these methods include their capability to adapt to (2)
dt r cos ϕ
Downloaded by 93.179.90.176 on September 16, 2019 | http://arc.aiaa.org | DOI: 10.2514/1.A34450
ω2e r cos ϕ sin ϕ sin ψ In Eqs. (14) and (15), αmax and βmax are the attack-angle and the
Cψ 2ωe sin ϕ − cos ψ tan γ cos ϕ (11) sideslip-angle limits, respectively, which are obtained by analyzing
V cos γ
the aerodynamic parameters of the grid wing; most of the time, these
X, Y, and Z in Eqs. (4–6) are aerodynamic drag, lift, and yaw terms are equal. In addition, ts is the standard trajectory stop guidance
accelerations (not forces), respectively. Note that, during the RCV time, and t is the current guidance time. In this way, the curves of α
atmosphere reentry, there is a rocket-bottom alignment velocity and β will be diagonal lines that change over time and are truncated at
vector. Therefore, angle of attack α and sideslip angle β are defined as αmax and βmax .
the angle between the rocket bottom (not the rocket head) and the
velocity vector, and thus, aerodynamic acceleration is as follows: X is III. New High-Precision Predictive Entry Guidance
aligned in the direction of the center of mass to the rocket bottom; Y is Method
perpendicular to X in the normal plane; and X, Y, and Z constitute a
right-hand coordinate system (shown in Fig. 2). To improve the guidance accuracy, it is necessary to decrease the
The aerodynamic acceleration equations are methodical error between the onboard model and the actual flight
conditions. Thus, the adaptive predictive entry guidance is divided
2 3 2 3 into two phases: an onboard adaptive aerodynamic fitting phase and a
X CD qS predictive-guidance phase. The research framework of this algorithm
6 7 16 7
4 Y 5 4 CL αqS 5 (12) is shown in Fig. 3.
m
Z CL βqS The method of fitting the deviation phase is designed to calculate
the aerodynamic acceleration deviations K D and KL (which are
defined as lift and drag acceleration deviations, respectively)
in which the dynamic pressure q is defined as
onboard. The predictor–corrector phase uses the coefficient to revise
Downloaded by 93.179.90.176 on September 16, 2019 | http://arc.aiaa.org | DOI: 10.2514/1.A34450
−βK t − ts if jβj ≤ jβmax j
β (15)
βmax if jβj ≥ jβmax j
Fig. 2 Landing and Body coordinate system. Fig. 3 Adaptive aerodynamic fitting method research framework.
4 Article in Advance / BOJUN, ZHANCHAO, AND GANG
8
Table 1 Deviation values >
< GV −gr sin γ− gϕ cos γ cos ψ
>
Gγ Vr − gr cosV γ gϕ sin γ cos ψ
2
Parameter Values (19)
ρ 20% >
>
: Gψ V cos γ sin ψ tan ϕ gϕ sin γ cos ψ
CD 20% r
CL 20%
m, kg 10
V, m∕s 50 3) In Eq. (18), it can be found that the drag and lift aerodynamic
X, m 5000
acceleration errors are only related to coefficients KD and KL .
Therefore, Eq. (16) can be substituted into Eq. (18), which simplifies
the calculation of the deviation coefficient.
Table 2 Guidance binding parameters In this section, the adaptive aerodynamic fitting method under the
current state is derived. In the actual calculation, the coefficients are
Parameter Values fit in each step and used as the entire entry phase deviation.
ρ 1.225e–r∕7717
CD 0.6 B. RCV Entry Predictive-Guidance Model
CL 0.3
m, kg 32,677 In Eq. (12) and Fig. 3, it is apparent that the control values are α and
αmax and βmax , rad 0.1 β, which are determined by coefficients αk and βk . Obviously, these
two coefficients will affect the longitude and latitude of the landing
point, respectively; thus, the linear iteration method cannot be used to
determine them. Therefore, their values can be iterated by calculating
Downloaded by 93.179.90.176 on September 16, 2019 | http://arc.aiaa.org | DOI: 10.2514/1.A34450
Substituting Eqs. (4–6) into Eq. (17) and ignoring acceleration due to
the rotation of Earth, Eq. (18) is derived:
8
>
< V i −X m GV V i1
(18)
: cos γ i V Gγ cos ψ i − V cosγ Gψ cosγ i1 cosψ i1
Ym Zm
>
2 3 2 3
∂θ ∂θ θf − θα θf − θβ In Eq. (21), dαk and dβk are the increments of coefficients αk and
6 ∂α ∂β 7 6 Δα ΔβK 7 βk . These equations can be solved using the elimination method.
6 K7 6 7
JαK ΔαK ; βK ΔβK 6 K 76
K
7 Because Eqs. (1–6) are nonlinear, Eq. (21) may have to be calculated
4 ∂ϕ ∂ϕ 5 4 ϕf − ϕα ϕf − ϕβ 5 two or three times before Δθ and Δϕ achieve their required
∂αK ∂βK ΔαK ΔβK accuracies.
(20)
in which θf and ϕf are the longitude and the geocentric latitude,
respectively, at the termination of the entry phase; and θα , ϕα , θβ , and IV. Evaluation via Monte Carlo Simulation
ϕβ are the calculated partial derivatives. The phase endpoint is A. Guidance Binding Parameters and Deviation Value
predicted once using the current αk and βk , and the errors of the phase
endpoint longitude and the geocentric latitude are written as Δθ and To best simulate the environment and the possible deviations of the
Δϕ. Then, the simultaneous equations are shown as Eq. (21): actual flight, the main function uses relatively accurate atmospheric
parameter interpolation tables to calculate the current state. Multiple
2 3 deviations are introduced for the combined calculation, as shown in
∂θ ∂θ " # " #
6 ∂α 7 Table 1, in which V and X are the initial velocity and the position
6 K ∂βK 7 dαK Δθ
6 7 (21) deviation, respectively, which are random values within a given
4 ∂ϕ ∂ϕ 5 dβK Δϕ interval and have random directions. Then, we obtain new
∂αK ∂βK r; θ; ϕ; V; γ; ψ as initial values.
Following Eq. (16), the guidance binding parameters are shown in V. Conclusions
Table 2. In this paper, a high-precision adaptive predictive entry guidance
In Table 2, the binding parameters are very simple; thus, there must method for the vertical recovery of an reusable core vehicle (RCV)
be a large error during the entry phase. The parameters αmax and βmax
is presented. Based on the characteristic of predictive algorithms
are estimated values that limit α and β.
that the onboard model can be changed easily, the dynamic model
of predictive guidance is modified to make the method suitable for
B. Result of the Adaptive Aerodynamic Fitting Method RCV vertical recovery. In addition, an adaptive aerodynamic fitting
Under the effects of various deviations, the measured and fitted method is added to decrease the discrepancies between the onboard
deviations of a trajectory are shown in Fig. 3. model and the actual environment before the predictive phase. The
KD and K L are measured deviations. In Fig. 4, it is apparent that fitting method used the output value of inertial measurement unit as
coefficients K Dm and KLm are produced as straight lines twice. This is input directly. Thus, a filter is needed to decrease the measurement
because, in the beginning, the RCV flies high and atmospheric error, which will be calculated into the KD and KL . The simulation
density is very low, leading to very few aerodynamic forces. results demonstrate that the algorithm is very robust and provides
Therefore, the predicted velocity at each step is not much different better accuracy than the traditional prediction guidance algorithm.
from the measured velocity, for which the adaptive aerodynamic The deviation analysis shows that the adaptive predictive entry
fitting method does not work. At the end of the flight, stop guidance guidance method is suitable for the two methods of RCV vertical
time ts was reached, which is designed to prevent deviation near the recovery.
end of the flight due to a high attack angle, which affects the flight The development of an entry guidance algorithm that uses a
stability. numerical predictor–corrector algorithm to develop a reference
Figure 5 shows that, when the adaptive aerodynamic fitting trajectory is promising. The problem of guidance deviation caused
method is working, the difference between the measured and fitting by then the conformity between the binding model and the actual
deviations is small and the algorithm achieves the expected results.
flight environment is solved by adding the onboard aerodynamic
acceleration fitting method before the predictive-guidance calculation.
C. Results of the Monte Carlo Simulation With the adaptive aerodynamic fitting method and the new iteration
To verify the effectiveness of the method proposed in this paper, method, when both α and β are added, even though the required
Fig. 6 shows the trajectory of the RCV vertical recovery at the launch computational time and resources are more than the requirements of a
site, which is under a launch coordinate system because γ is steep, and typical predictor–corrector guidance method, the adaptive predictive
the latitude and longitude change only slightly. entry guidance method has better precision.
Figure 6 shows that, with the guidance method, the RCV flies
directly toward its intended landing point. In Fig. 7, α and β are shown
during entry. It is apparent that α and β gradually decrease as height
decreases, and are visibly limited by amplitude in the upper References
atmosphere. [1] Horvath, T. J., Aubuchon, V. V., Rufer, S., Campbell, C., Schwartz, R.,
The results of 1000 Monte Carlo simulations are shown in Fig. 8 as Mercer, C. D., Tack, S., Spisz, T. S., Gibson, D., Osei-Wusu, K., et al.,
the X and Z direction position deviations under the launch coordinate “Advancing Supersonic Retro-Propulsion Technology Readiness:
system. Infrared Observations of the SpaceX Falcon 9 First Stage,” 2017 AIAA
SPACE and Astronautics Forum and Exposition, AIAA Paper 2017-
The deviation analysis of the shutdown point along the X and Z
5294, Sept. 2017,
axes is shown in Table 3. The deviation of the final shutdown point is doi:10.2514/6.2017-5294
very small, and a meter-scale guidance precision can be attained, [2] Lu, P., “Propellant-Optimal Powered Descent Guidance,” Journal of
which is better than a typical predictor–corrector entry guidance Guidance, Control, and Dynamics, Vol. 41, No. 4, 2017, pp. 813–826.
algorithm. doi:10.2514/1.G003243
[3] Lu, P., “Adaptive Powered Descent Initiation and Fuel-Optimal
Guidance for Mars Applications,” 2018 AIAA Guidance, Navigation,
Table 3 Deviation analysis and Control Conference, AIAA SciTech Forum, AIAA Paper 2018-
0616, Jan. 2018.
Recovery at launch site Recovery on ship doi:10.2514/6.2018-0616
X Z X Z [4] Wang, T., Zhao, H. B., Zeng, L., and Tang, G. J., “A Robust Predictor-
Corrector Entry Guidance,” Aerospace Science and Technology,
Mean value –9.9 15.7 13.8 –10.6
Vol. 66, July 2017, pp. 103–111.
Variance 1.7 21 39.6 56.3
doi:10.1016/j.ast.2017.03.010
Article in Advance / BOJUN, ZHANCHAO, AND GANG 7
[5] Lu, P., “Entry Guidance: A Unified Method,” Journal of Guidance, [9] Kozynchenko, A. I., “Analysis of Predictive Entry Guidance for a Mars
Control, and Dynamics, Vol. 37, No. 3, 2014, pp. 713–728. Lander Under High Model Uncertainties,” Acta Astronautica, Vol. 68,
doi:10.2514/1.62605 Nos. 1–2, 2011, pp. 121–132.
[6] Kluever, C. A., “Entry Guidance Performance for Mars Precision doi:10.1016/j.actaastro.2010.08.005
Landing,” Journal of Guidance, Control, and Dynamics, Vol. 31, No. 6, [10] Miele, A., Zhao, Z. G., and Lee, W. Y., “Optimal Trajectories for the
2008, pp. 1537–1544. Aeroassisted Flight Experiment. Part 1: Equations of Motion in an
doi:10.2514/1.36950 Earth-Fixed System,” NASA CR-186134, 1989.
[7] Casoliva, J., Lyons, D. T., Wolf, A. A., and Mease, K. D., “Robust doi:10.1016/0094-5765(90)90116-3
Guidance via a Predictor-Corrector Algorithm with Drag Tracking for [11] Shen, Z. J., and Lu, P., “On-Board Generation of Three-Dimensional
Aero-Gravity Assist Maneuvers,” AIAA Paper 2008-6816, Aug. 2008. Constrained Entry Trajectories,” Journal of Guidance, Control, and
doi:10.2514/6.2008-68s16 Dynamics, Vol. 26, No. 1, 2003, pp. 111–121.
[8] Lafleur, J. M., “The Conditional Equivalence of ΔV Minimization and doi:10.2514/2.5021
Apoapsis Targeting in Numerical Predictor-Corrector Aerocapture
Guidance,” NASA, Johnson Space Center TM-2011-216156, Houston, C. Bonnal
TX, Aug. 2011. Associate Editor
Downloaded by 93.179.90.176 on September 16, 2019 | http://arc.aiaa.org | DOI: 10.2514/1.A34450