See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/360652959

Comprehensive Load Relief of Launch Vehicle with the Constraints of Legacy

Stages

Article  in  AIAA Journal · August 2022

DOI: 10.2514/1.J061610

CITATIONS READS
0 118

6 authors, including:

Zhengyu Song
Zhejiang University
63 PUBLICATIONS   297 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Zhengyu Song on 17 May 2022.

The user has requested enhancement of the downloaded file.

 Comprehensive Load Relief of Launch Vehicle with the
Constraints of Legacy Stages

Zhengyu Song ∗
China Academy of Launch Vehicle Technology, Beijing, China, 100076
College of Control Science and Engineering, Zhejiang University, Hangzhou, China, 310027

Hao Pan†
Beijing Aerospace Automatic Control Institute, Beijing, China, 100854

Shanshu Xu‡ ,Jinglin Li§ , and He Yuan¶

Beijing Institute of Astronautical Systems Engineering, Beijing, China, 100076

Cong Wang‖
Beijing Aerospace Automatic Control Institute, Beijing, China, 100854

Facing diversified launch needs, it is uneconomical to re-develop new rockets from scratch.

The concept of combining legacy modules into a new launcher has become attractive. However,

these mature modules may not adapt to new flight profiles due to differences in factors such

as the propulsion system configuration, payloads, and trajectories. Thus, structural load

relief measures play a key role in module integration. A comprehensive load relief strategy

applied in LM-8 is introduced, including the inflight load reduction by engine throttling, wind

biasing trajectory, and onboard load relief control techniques. A unified analysis process

for the elastic loads caused by gusts and fluctuating pressures was proposed, and an integral

optimization problem for simultaneous planning with complicated constraints, such as the

maximum dynamic pressure, wind biasing, and throttling level and time, are discussed. A real-

time load relief technology based on an extended state observer, which predicts the angular

acceleration caused by the wind, is proposed. Its efficiency, adaptability, influence on the

stability and control accuracy, and its application in the LM-8 maiden flight, are studied. The

proposed scheme expands the launch probability of LM-8 and provides a systematic solution

for the load relief design and control of in-service rockets.

∗ Professor,China Academy of Launch Vehicle Technology and Zhejiang University, songzhengyu@zju.edu.cn (Corresponding Author).
† Dr., Beijing Aerospace Automatic Control Institute, hpan@amss.ac.cn.
‡ Dr., Beijing Institute of Astronautical Systems Engineering, shanshuxu@sina.com.
§ Dr., Beijing Institute of Astronautical Systems Engineering, 1107410436@qq.com.
¶ Mr., Beijing Institute of Astronautical Systems Engineering, yuanhe0823@163.com.
‖ Dr., Beijing Aerospace Automatic Control Institute, wangcong0404@126.com.
 I. Introduction
As China’s new generation launch vehicles (NGLVs) LM-7 and LM-5 begin to enter the stage [1, 2], a medium-lift

launcher for sun synchronous orbit (SSO) launching is needed to supplement this family of NGLVs. However, it is

uneconomical to develop such a launcher from scratch. Therefore, it is highly attractive to combine the modules of

in-service long march launch vehicles (LMLVs) into a new rocket. LM-8 was born under this background [3].

The SSO (700 km) performances of the family of long march launchers was no more than 3 tons for the existing

configurations. However, we found that by only combining the side and core boosters of LM-7 and the 3rd stage of

LM-3A into a new launcher, called LM-8, the performances can be extend to 5 tons. Although this solution could

greatly reduce the development cycle and costs, whether the legacy hardware can adapt to the new flight profile is

the main concern. The second stage of LM-8 is inherited from the last stage of LM-3A, whose bending moment is

unsuitable for the new mission profile due to the greater endo-atmospheric flight overload and the larger and longer

fairing. If no measures are taken, the second stage should be re-designed, and the objective of the module combination

cannot be achieved.

The structural loads are mainly related to the axial load and bending moment. The greater the acceleration is, the

greater the axial load and dynamic pressure become. The larger and longer the fairing is, the greater the bending

moment experienced by the second stage becomes. The bending moments that the stages experience usually include a

quasi-static component caused by high-altitude wind, which is characterized by 𝑞𝛼 , and a dynamic component caused

by elastic loads. 𝑞𝛼 is the product of the dynamic pressure 𝑞 and the angle of attack (AOA) 𝛼 , which is a simplified

form of 𝑞 sin 𝛼 ( sin 𝛼 ≈ 𝛼 if 𝛼 is small), representing the aerodynamic pressure acting on the launcher perpendicular to

the longitudinal axis of the body. Vause and Starr proposed a trajectory-based load analysis method, and each point on

the trajectory was treated as a load case [4]. Starr et al discussed a flexible body coupled load assessment of the day of

launch, which quantified the flight loads in terms of the structural load components rather than the typically used 𝑞𝛼

metric [5]. In Refs. [6] and [7], load exceedances at several locations were found during Space Launch System (SLS)

Design Analysis Cycles, and a maximum load is then set for launches.

Load relief is scheduled to solve the load exceedance cases, which has been studied in detail. A survey of the many

publications shows that only one aspect of the problem has been treated at a time [8, 9]. Engine throttling is an effective

way to alleviate the load, so as to reduce the speed when flying through the transonic or maximum dynamic pressure

regimes, just as the space shuttle main engines throttle down [10]. Betts and Frederick reviewed six technical means

to realize throttling of a liquid engine [11]. One approach was a high-pressure drop injection system with propellant

flow regulation, which is the solution used for the throttling of YF100 engines. However, the throttling will affect the

carrying capacity, and the degree of throttling is not unlimited.

Use of wind biasing trajectories is another method of load alleviation [12]. The wind biasing trajectory aims to

eliminate the additional AOA generated by a steady wind. This additional AOA increases the aerodynamic load acting

2
on the rocket, leading to complex unsteady flow in the transonic regime and increased buffet loads therefrom. With this

method, the trajectory is shaped such that the vehicle flies with a prescribed AOA profile in the presence of a wind

field during the high-dynamic-pressure regime. Since 1990, an onboard “load relief guidance” mode has been adopted

in the Delta rocket, where the wind “curve fits” from a T-minus 2 h wind profile, not the trajectory, were loaded into

the guidance computer approximately 10 min before liftoff [13, 14]. The wind profile was modeled using a low-order

polynomial curve fit or a wind table, and an AOA estimation and feedback system was integrated into the guidance

control [15]. A more detailed discussion of these wind models can be found in Ref. [16].

In 1992 and 1995, China’s LM-2E Y4 and Y6 suffered two failures during winter launches, which were determined

to be related to high-altitude winds, causing breaks of the fairings and satellites when flying through the transonic phase.

As a result, the design specifications of LM rockets were re-examined, and for the first time, the wind biasing technique

was applied during LM-2E Y7 and Y8 launches of the Asia II and Echo I satellites in 1995 [17].

The first study of wind biasing techniques in Europe for trajectory optimization was presented in Ref. [18].

Nowadays, many vehicle guidance systems have moved away from using monthly mean wind profiles as the input

and instead use the real-time measured wind profile as the input of their tilt programs [19]. However, the effect of

wing biasing or load relief guidance depends on responsive measurements or accurate prediction of the wind profiles.

Moreover, SSO launches at the Wenchang satellite launch center face predominant west-to-east winds, and thus, the

tailwind effect or simply aligning the rocket’s longitudinal axis along the relative velocity vector would dramatically

affect the performance.

Finally, a third method, active onboard load relief control, has been proposed to treat the wind disturbances in real

time [20]. Two different kinds of sensors– accelerometers and AOA meters– have been used for the control process,

providing equivalent rigid body responses [21]. The Load Relief control based on lateral Acceleration (LRA) is the

most common method [22, 23]. Other methods include tracking and estimating the wind velocity and its disturbances

to guidance commands thereafter with a robust wind disturbance observer [24], using the feedback of the deviation

between the attitude angle rate and the command rate to reduce the AOA [25], and adopting a nonlinear controller for

load alleviation and balancing its effect on the tracking accuracy during different flight phases [26]. In the Ares I flight

control system, an anti-drift (AD) algorithm including force and moment balances with input data from rate gyros and

accelerometers was proposed, where the external forces and moments could be caused by aerodynamics, thrust vector

control (TVC) offsets, gusts, and winds [27]. Orr reviewed the load relief control techniques, and proposed that novel

methods using modern strapdown GPS/INS combined with near real-time measurements on the winds can yield very

good performance, but they also rely on the availability of reliable AOA data [28]. However, onboard load relief control

will have an impact on the flight stability and control accuracy, and its effect is closely related to the features of the

rockets and the wind field. Thus, load relief control is still based on design knowledge and experience, leading to a

variety of techniques being developed and continuous improvements.

3
 Due to the limitations of these methods and their impacts on the carrying capacity and stability, the effect of each

method on load relief cannot solve the problems faced by the LM-8. Thus, the contributions of this paper are as follows.

A comprehensive load relief scheme is systematically proposed, and an integral problem covering the engine throttling

and wind biasing process is established to optimize the 𝑞𝛼 constraint. An innovative method for onboard Load Relief

control based on an Extended state observer (LRE) is presented, which predicts the angular acceleration caused by the

wind and compensates the disturbance moment therefrom, and its load alleviation effect was first demonstrated in the

maiden flight of LM-8.

The remaining contents of this paper are arranged as follows. The basic features of the LM-8 rocket, the load relief

roadmap, and the elastic load analysis process are summarized in Section 1. Section 2 discusses the wind biasing

trajectory planning. By making 𝑞𝛼 as small as possible while simultaneously meeting other constraints, enough load

margins are retained to counter the disturbances, uncertainties, and wind field changes. In Section 3, the LRE is analyzed

and compared with the LRA. Its physical mechanism, the factors affecting its efficiency, the adaptability to rocket

features, and the influences on the flight stability and control accuracy are studied in detail. Simulations and flight

results are discussed in Section 4, and Section 5 concludes the paper.

II. Comprehensive Load Relief Measures

A. Overview of LM-8 Rocket

The diagram of the LM-8 is shown in Fig. 1. It is a two-stage, medium-lift, cryogenically fueled launch vehicle with

two side boosters. The core stage has a diameter of 3.35 m and is equipped with two YF100 engines, each of which can

perform a two-way swing. The side booster has a diameter of 2.25 m and is equipped with one YF100 engine that can

perform a one-way swing. The second stage has a diameter of 3 m and is equipped with two YF75 engines. The YF100

engines use liquid oxygen (LOX)/kerosene as the propellant, while the YF75 engines use LOX/LH2 as the propellant.

The diameter of the inter-stage section transitions from 3.35 to 3 m, which is a newly developed module. The fairing has

a diameter of 4.2 m, and a length of 12 m. LM-8 is cost effective and capable of carrying 5 tons for SSO missions (700

km).

The acceleration of LM-8 flying in the atmosphere is larger than that of LM-3A, increasing the dynamic pressure.

The fairing is also longer and wider than those used in the LM-3A series rocket, increasing the pressure on the fairing

and the arm of the force acting on the 2𝑛𝑑 stage. These caused the moment that the last stage of LM-8 experienced in

the transonic regime to be 104% greater than that experienced during the LM-3B flight profile, causing the structural

load to exceed the tolerance of this legacy stage.

4
 Fairing with a
diameter of 4.2 m

2nd stage inherited

from the last stage
of LM-3A series

Newly designed
inter-stage
1. Fairing
2. Satellites
3. LV/Payload adapter
4. Instrument cabin
5. LH2 tank of 2nd stage
Side and core boosters 6. LOX tank of 2nd stage
inherited from LM-7 7. YF-75 engines
8. Inter-stage section
9. LOX tank of 1st stage
10. Inter-tank section
11. Kerosene tank
12. YF100 engines of 1st stage
13. Nose-cone of side booster
14. YF100 engines of side booster

Fig. 1 Configuration of LM-8.

B. Elastic Load Analysis

After calculating the dynamic bending moments, the quasi-static bending moment can then be determined by

subtracting the dynamic components from the stage’s bearing capacity. The dynamic bending moments, or the elastic

loads, are caused by gust and buffet forces.

Load analysis usually employs linearization of the flow-structure interactions around a prescribed reference flight

condition to significantly simplify the problem, thereby greatly reducing the computational burden [29]. The turbulence

is regarded as a set of component velocities superimposed on the background steady flow. The well-established methods

for elastic load calculations, which are based on linear aerodynamic models, are solved in the frequency domain, and the

input and output of the system have the following relationship:

𝑿 (Ω) = 𝑯 (Ω) 𝑭 (Ω) , (1)

where 𝑭 (Ω), 𝑿 (Ω), 𝑯 (Ω) represent the excitation vector, the response vector, and the frequency response function

matrix between them, respectively.

It is difficult to carry out Fourier transforms for loads such as gust loads and buffet loads, as they both exhibit random

characteristics. Thus, random vibration theory and related statistical theory are applied for elastic load analysis.

5
 The elastic load analysis begins by solving for 𝑞 𝑗 , the 𝑗-th generalized displacement, under the disturbances of a

gust or the transonic fluctuating pressure, which is then multiplied by the modal force to obtain the elastic loads on the

rocket. For a gust, 𝑞 𝑗 is related to the forces on all the nodes of the rocket, whose amplitudes are determined by the

aerodynamic coefficients with the same frequencies. For the fluctuating pressure, 𝑞 𝑗 is related to disturbances only near

the shoulder of the rocket fairing and the side boosters, whose frequencies and amplitudes differ from each other. Thus,

the coherence functions between each action point should be considered.

A unified process of elastic load analysis is summarized below. The maximum value of 𝑞 𝑗 with a 99% probability,

which is used to calculate the elastic loads, is determined by 𝐶𝑔 and 𝐶 𝑝 based on the probability distribution model.

Process for elastic load analysis:

1) Establish the open-loop motion equation of the generalized elastic displacement, 𝑞 𝑗 , in the modal domain.

For the gust, the simplest open-loop structural dynamics equation is as follows [30]:

𝐹 𝑗 (𝑡)
𝑞¥ 𝑗 (𝑡) + 2𝜉 𝑗 𝜔 𝑗 𝑞¤ 𝑗 (𝑡) + 𝜔2𝑗 𝑞 𝑗 = , (2)
𝑀𝑗

where 𝜔 𝑗 , 𝜉 𝑗 represent the 𝑗-th natural frequency and the modal damping ratio of the system, respectively; 𝐹 𝑗

denotes the 𝑗-th-order generalized force generated by gust, and 𝑀 𝑗 denotes the 𝑗-th modal mass.

For the fluctuating pressure disturbances, the simplest open-loop structural dynamics equation is

1 Õ𝑘
𝑞¥ 𝑗 (𝑡) + 2𝜉 𝑗 𝜔 𝑗 𝑞¤ 𝑗 (𝑡) + 𝜔2𝑗 𝑞 𝑗 (𝑡) = 𝑁 𝑛 𝑗 (𝑡), (3)
𝑀𝑗 𝑛=1

where 𝑁 𝑛 𝑗 denotes the 𝑗-th-order generalized external force, and 𝑘 denotes the number of fluctuating pressures

acting simultaneously on the rocket.

2) Solve for the transfer function 𝐻 (𝑖Ω) = 𝑞 𝑗 (𝑖Ω)/𝐹 𝑗 (𝑖Ω) for a gust load or 𝐻 (𝑖Ω) = 𝑞 𝑗 (𝑖Ω)/𝑁 𝑛 𝑗 (𝑖Ω) for the

fluctuating pressure load based on the Laplace transforms of Eq. (2) or Eq. (3), respectively.

3) Calculate the power spectral density (PSD) of 𝐹 𝑗 : Φ𝐹 (Ω) or that of 𝑁 𝑛 𝑗 : Φ 𝑁 (Ω).

4) For linear systems, the responses to continuous turbulence may be calculated entirely in the frequency domain

with a root mean square response [31]:

∫ ∞
𝑞¯ 2𝑗 = |𝐻 (𝑖Ω)| 2 Φ𝐹 (Ω) 𝑑Ω, (4)
0

or
∫ ∞
𝑞¯ 2𝑗 = |𝐻 (𝑖Ω)| 2 Φ 𝑁 (Ω)𝑑Ω, (5)
0

where 𝑖 is the imaginary unit, and 𝑞¯ 2𝑗 denotes the mean square value of 𝑞 𝑗 .

6
 5) Calculate the maximum value with a 99% probability and a 90% confidence level:

𝑞 𝑗 max = 𝐶𝑔 𝑞¯ 𝑗 for a gust load or 𝑞 𝑗 max = 𝐶 𝑝 𝑞¯ 𝑗 for a fluctuating pressure load, where 𝐶𝑔 and 𝐶 𝑝 are constant

coefficients.

If a launch vehicle were subjected to a large number of actual wind profiles, these gust profiles would yield loads that

are equivalent to some statistically conservative level [30]. The PSD of the gust, Φ𝐹 (Ω), can be analytical. At present,

the Dryden [32] and von Kármán [31, 33] models are generally used to describe the gust spectrum, and the Dryden

model was selected for gust load analysis for LM-8. The PSD of the fluctuating forces, Φ 𝑁 (Ω), is usually obtained after

processing wind tunnel experiment results based on the flow similarity principle [34]. This topic is beyond the scope of

the paper, and more detailed discussions about the numerical simulation, measurement, and analysis of buffet loads on

the fairings can be found in Refs. [35] and [36] and the references therein.

Currently, systematic and cost-effective approaches are being sought to develop models capable of describing both

linear and nonlinear effects for a range of cases [37]. However, LM-8 still uses the linearization method. A more

detailed discussion about the total load integration from sources such as the dynamic pressure, gusts, buffet loads, and

thrust oscillation and vectoring can be found in Ref. [38].

After the elastic loads are determined, the quasi-static bending moment should also be determined too by subtracting

the elastic loads from the structure bearing capacity, and finally, the limit of 𝑞𝛼 for the LM-8 was determined as 1700

Pa·rad.

C. Measures Considered for Load Relief

To reduce 𝑞𝛼 , engine throttling is scheduled inflight. The diagram of the YF100 engine is shown in Fig. 2. A

flow regulator is in the feed line of the gas generator. When throttling, the motor in the regulator rotates to reduce the

opening of the regulator. The fuel flowing to the generator and the combustion temperature of the generator are reduced

therefrom. In turn, the output power of the turbine is reduced, and the flow of LOX and kerosene into the engines also

decreases, reducing the thrust [39]. A small action of the regulation device can produce a significant change of the

turbine power, which can lead to a wide variation of the thrust with only a small deviation of the mixture ratio.

The engines operate over a long period under low-operation conditions when throttling, so it is necessary to consider

the adaptability of the engine structural layout, the engine performance, the evaporator, the residual axial force of the

turbo oxygen pump, and other engine components. Thus, the throttling degree is usually limited. For YF100 engines,

throttling to 75% is considered a safe condition, although the actual limit is far beyond that.

A wind biasing trajectory is also needed. However, the complete upwind flight of the rocket will deviate from

the flight path, resulting in the loss of performance and the difficulty of selecting the debris landing area. Then, an

in-flight real-time load relief method is called, such as LRA, which is widely applied. The real-time load relief effect

was first included in a go/no-go decision for the LM rockets during the maiden flight of LM-8. Furthermore, LRE is

7
 LOX LOX
preloading
Kerosene Generator turbine Generator
fuel valve pump fuel valve
Oxygen-enriched gas Helium
High temperature gas heater
LOX
Igniter main
valve
(2)
Turbine Ignition
conduit
Gas
generator

Thrust
chamber
LOX
main
pump

Fuel
main Initiating box
valve
Throttle
valve
1st fuel pump
Drain valve
2nd fuel pump
Flow regulator

Fig. 2 System diagram of YF100 engines.

proposed to counter the changes of the wind field or the shear wind, whose effect is still taken as the design margin. The

comprehensive load relief measures adopted by LM-8 are summarized in Fig. 3, and the key technologies are described

below.

Design constraints Design input

Trajectory, payload, engine features, mass Launch day
Impact on Statistical wind fields wind field
properties, loads constraints,
carrying (altitude, velocity,
environmental parameters, jettison falling
capacity direction, frequencies)
area limits, TC/TM requirements.
Throttling
depth
② Engine ① Wind
throttling Requirement
Off-line trajectory planning
biasing
Reference Altitude, velocity,
trajectory mass, center of
General parameter gravity, pressure
center
Theoretical diagram, aero
③ Structural analysis Go/no-go
features & distributions, decision
specializing in flexible features
tank pressuring.

,
Load relief based ④ Tracking guidance and Load relief
on acceleration onboard load relief control based on ESO

Fig. 3 Comprehensive load relief solution.

8
 III. Wind biasing Trajectory Design

A. Principle of Wind Biasing

The principle of wind biasing is shown in Fig. 4. 𝑂 − 𝑥 𝑏 𝑦 𝑏 𝑧 𝑏 is the launcher body coordinate system, where 𝑥 𝑏

represents the longitudinal axis of the launcher, 𝑦 𝑏 points along the normal direction, and the launcher flies within the

𝑂 − 𝑥 𝑏 𝑦 𝑏 plane. 𝑧 𝑏 is perpendicular to the 𝑂 − 𝑥 𝑏 𝑦 𝑏 plane and satisfies the right-hand rule. The left side is a diagram

of the AOA, and the right side defines the side slip angle, 𝛽. The following analysis is based on the AOA, and that for

sideslip angle is similar.

Reference base

Fig. 4 Diagram of wind biasing.

When there is no wind, the air flow is stationary relative to the ground, and the wind speed 𝑉®𝑤 = 0. Thus, 𝑉®𝑘 , the

ground velocity (velocity of rocket relative to the ground), coincides with the airspeed 𝑉®𝑎 . 𝛼 is just 𝛼 𝑘 , which represents

the AOA of ground velocity, also known as ballistic AOA; and

𝜙 = 𝜃 + 𝛼𝑘 , (6)

where 𝜙 is the pitch attitude angle, and 𝜃 denotes the trajectory inclination.

In the case with wind, 𝑉®𝑎 is determined by the following equation:

𝑉®𝑎 = 𝑉®𝑘 − 𝑉®𝑤 . (7)

At this time, the AOA is the angle between the longitudinal axis 𝑥 𝑏 and 𝑉®𝑎 , which is equivalent to introducing an

additional wind angle of attack, 𝛼 𝑤 ,

𝛼 = 𝛼𝑘 + 𝛼𝑤 , (8)

9
 To apply wind biasing, 𝑥 𝑏 is then regulated to 𝑥 0 𝑏 by adjusting the attitude as follows:

𝜙 = 𝜃 + 𝛼𝑘 − 𝛼𝑤 . (9)

The ballistic AOA is updated as 𝛼 𝑘 0 = 𝛼 𝑘 − 𝛼 𝑤 , and the total AOA is

𝛼 0 = 𝛼 0 𝑘 + 𝛼𝑤 = 𝛼𝑘 . (10)

By comparing Eq. (8) and Eq. (10), 𝛼 𝑤 is eliminated from the total AOA.

The ballistic AOA (𝛼 𝑘 ) remained, which could help the rocket to change the velocity direction in the pitch channel to

complete the gravity turn. It is not a constant value, and can be fitted by empirical (exponents or polynomials) equations.

Yu introduced the following exponential expression of 𝛼 𝑘 [17]:

𝛼 𝑘 = 4𝛼0 𝑒 𝛼1 (𝑡1 −𝑡) (𝑒 𝛼1 (𝑡1 −𝑡) − 1), (11)

where 𝑡1 represents the moment when wind biasing begins, 𝛼0 represents the ballistic inclination when the first stage

shuts down, 𝛼1 determines the time when the extreme value is reached. 𝛼 𝑘 would reach the negative extremum at a very

fast rate, after which its absolute value would decrease and approach 0 at a prescribed exponential rate. For example, 𝛼 𝑘

would be almost 0 in the transonic and maximum dynamic pressure regimes.

B. Trajectory Planning

In this section, an integrated trajectory planning problem, including dynamics, constraints, and objectives, is

formulated. The wind biasing and throttling concerns are included in the dynamics, and the jettison landing area

limitations are set as constraints. Different from traditional trajectory optimization problems, the objective is not to

maximize the performance to reach orbit but to minimize the 𝑞𝛼 values, while keeping the payload mass as the terminal

constraint. The flight during the boost phase, the first stage, the first burn of the second stage, coasting, and the second

burn are denoted by 𝑖 = 1, 2, 3, 4, 5, respectively. To save space, only the models during the boost phase are explained in

detail, and the equations for other phases are similar and omitted. However, the differences in other phases are pointed

out accordingly.

• Dynamics

Due to the relatively short range of the ascent flight, the influence of the Earth’s curvature changes and the effect of

the Earth’s rotation on the gravitational acceleration can be ignored. By considering a flight when the boosters are

10
involved, the following simplified dynamic models are derived:

𝑽¤ = 𝑾
¤ + 𝒈, 𝑷¤ = 𝑽, 𝑚¤ 1 = −𝑑𝑚 1 , 𝑚¤ 2 = −𝑑𝑚 2 , 𝑚¤ 3 = 0,
 𝑇
¤ 1
𝑾 = 𝑚1 +𝑚2 +𝑚3 𝑮 𝐵 𝐹𝑎𝑥1 + 𝑇1 + 𝑇2 𝐹𝑎𝑦1 𝐹𝑎𝑧1
𝐼 ,



 𝐼𝑠 𝑝𝑖 𝑑𝑚 𝑖 , 𝑡 ≤ 𝑡 𝑡 ℎ𝑠 (12)






𝑇𝑖 = 𝑘 𝑇 𝐼𝑠 𝑝𝑖 𝑑𝑚 𝑖 , 𝑡 𝑡 ℎ𝑠 < 𝑡 ≤ 𝑡 𝑡 ℎ𝑒 , 𝑘 𝑇 ∈ [𝑘 𝑇 min , 1] , 𝑖 = 1, 2






 𝐼𝑠 𝑝𝑖 𝑑𝑚 𝑖 ,
 𝑡 > 𝑡 𝑡 ℎ𝑒


¤ 𝒈 and 𝑷 represent the velocity, the apparent acceleration, the gravity, and the position, respectively, defined
where 𝑽, 𝑊,

¤ 𝑑𝑚 represent the mass-flow-rate; 𝑚 1 , 𝑚 2 , 𝑚 3 , denote the masses

in the launching inertial coordinate system (LICS); 𝑚,

of side boosters and first and second stages, respectively; 𝑡 𝑡 ℎ𝑠 , 𝑡 𝑡 ℎ𝑒 represent the start and end times of engine throttling,

respectively; 𝐹𝑎𝑥1 , 𝐹𝑎𝑦1 , 𝐹𝑎𝑧1 are the aerodynamic forces. In the throttling state, the propellant flow and specific impulse

of each stage, 𝐼𝑠 𝑝𝑖 , will decrease accordingly. Here, the changes of these two parameters are no longer distinguished,

and the thrust variation is directly expressed by the factor 𝑘 𝑇 . The throttling can be realized by the engines of the side

boosters or the first stage, or both, which is determined according to the planning results.

The transformation matrix from the vehicle’s coordinate system to the LICS is as follows:

 
cos 𝜙 cos 𝜓
 𝑎 𝑎 − sin 𝜙 𝑎 cos 𝜙 𝑎 sin 𝜓 𝑎 
 
 
𝐺 𝐼𝐵 =  sin 𝜙 𝑎 cos 𝜓 𝑎 cos 𝜙 𝑎 sin 𝜙 𝑎 sin 𝜓 𝑎  , (13)
 
 
 − sin 𝜓 0 cos 𝜓 𝑎 
 𝑎
 

where 𝜙 𝑎 , 𝜓 𝑎 represent the pitch and yaw angles under LICS. During the wind biasing period, 𝜙 𝑎 and 𝜓 𝑎 are in the

form of Eq. (9).

The aerodynamic forces are as follows:

     
𝐹   cos 𝛼 cos 𝛽
 𝑎𝑥1   sin 𝛼 − cos 𝛼 sin 𝛽 𝐷 
 
     
     
 𝐹  = − sin 𝛼 cos 𝛽
 𝑎𝑦1   cos 𝛼 sin 𝛼 sin 𝛽  𝐿 ,
  (14)
     
     
𝐹  
 𝑎𝑧1   sin 𝛽 0 cos 𝛽  𝑍 
 
     

where 𝐷, 𝐿, 𝑍 represent the drag, the lift, and the lateral force, respectively. 𝛼 and 𝛽 are replaced by the design AOAs,

𝛼 𝑘 (Eq. (11)) and 𝛽 𝑘 , during the wind biasing period.

The flight of the first stage alone and the second stage have similar equations, but the aerodynamic force and engine

11
throttling are no longer considered, and the thrust of the coasting phase (𝑖 = 4) during the second stage is 0.

• Constraints

The booster phase is considered as an example, the constraints are shown as follows:


𝑽 (𝑡0 ) = 𝑽0 , 𝑷 (𝑡 0 ) = 𝑷0 ,











 𝑚 (𝑡 0 ) = 𝑚 10 + 𝑚 20 + 𝑚 30 ,


  0.5   3.15 (15)
¤ 𝐶𝑙 𝜌 k𝑽 k
≤ 𝑄¤ max ,



 𝑄 = √


 𝑅 𝑛 𝜌 0 𝑣 0



 𝑭 𝑽 𝑡
, 𝑷 𝑡
, 𝑚 𝑡

− [𝜆, 𝐵] 𝑇 ≤ Δ𝑅,

¯
𝑤ℎ𝑒𝑛 |𝐻| ≤ 𝐻.

 𝐷 1, 𝑓 1, 𝑓 1, 𝑓

The first two lines are the initial conditions of the velocity, position, and mass, and the initial conditions exist for

each stage. 𝑚 10 , 𝑚 20 , 𝑚 30 represent the initial mass of booster, first stage and second stage, respectively. 𝑄¤ is the heat

flow, considered only when flying in the atmosphere. 𝜌 and 𝑽 represent the current atmospheric density and velocity,

respectively, while 𝜌0 and 𝑣 0 denote the reference density and velocity, respectively; and 𝐶𝑙 is a constant coefficient

concerned with 𝜌0 and 𝑣 0 . 𝑅𝑛 is the nose radius. The last function, 𝑭𝐷 ,calculates the range of the landing points of the

side booster debris, where 𝜆, 𝐵 denote the longitude and latitude coordinates of planned landing points of debris, 𝐻 is

the altitude, 𝐻¯ is the altitude of the landing point, and Δ𝑅 is the allowable landing point deviation. This function is also

included for the landing constraints of the core stage and fairing.

In addition, the following constraints are imposed:


𝑁 𝑥1 = 𝑊¤ 𝑥1 /𝑔 < 𝑁 𝑥1 max ,








𝜙¤𝑎𝑖,min ≤ 𝜙¤ ≤ 𝜙¤𝑎𝑖,max ,

𝜙 𝑎𝑖,min ≤ 𝜙 ≤ 𝜙 𝑎𝑖,max ,








𝜓 𝑎𝑖,min ≤ 𝜓 ≤ 𝜓 𝑎𝑖,max , 𝜓¤ 𝑎𝑖,min ≤ 𝜓¤ ≤ 𝜓¤ 𝑎𝑖,max , (16)










 𝑚 𝑡𝑖, 𝑓 ≥ 𝑚¯ 𝑖, 𝑓



 𝑭𝐸 𝑽 𝑡5, 𝑓 , 𝑷 𝑡5, 𝑓 − 𝑎 𝑓 , 𝑒 𝑓 , 𝑖 𝑓 , Ω 𝑓 , 𝜔 𝑓 , 𝑓 𝑓 𝑇 ≤ 𝜺,




 


where 𝑖 = 1, 2, 3, 4, 5, 𝑁 𝑥1 is the axial overload, 𝑊¤ 𝑥1 is the axial apparent acceleration and 𝑚 𝑡 𝑖, 𝑓 , 𝑚¯ 𝑖, 𝑓 represent the

terminal mass and terminal mass constraint including safety margin of propellant. For multistage rockets, to ensure full

performance and ensure flight safety, the flight profile can be segmented according to the available propellant of each

stage, but the state variables are continuous, which means the values of the terminal velocity and position of the former

stage are the initial conditions of the current stage. However, the mass would undergo a sudden change when the stages

or fairing were separated. For the last stage, the payload mass is also included in 𝑚¯ 𝑖, 𝑓 . The function 𝑭𝐸 determines

12
the orbital elements when injection, including 𝑎 𝑓 , the semi-major axis; 𝑖 𝑓 ,the inclination; 𝑒 𝑓 , the eccentricity; Ω 𝑓 ,

the longitude of ascending intersection; 𝜔 𝑓 , the argument of perigee, and 𝑓 𝑓 , the true anomaly, respectively. 𝜺 is the

tolerance for orbit accuracy.

• Objective function

The following objective function is formulated:

min 𝐽 = max (|𝑞𝛼 (𝑡)|) , 𝑡 ∈ [𝑡 𝑠 , 𝑡 𝑒 ] , (17)

where 𝑡 𝑠 and 𝑡 𝑒 are the start and end times considering maximum dynamic pressure, which can also be prescribed in

advance. The minimization of 𝑞𝛼 can leave a sufficient margin to counter disturbances and uncertainties when deviating

from the nominal conditions.

In a summary, the state variables of the problem are [𝑽, 𝑷, 𝑚], and the control variables [𝜙 𝑎 , 𝜓 𝑎 ]. 𝑘 𝑇 , 𝑡 𝑡 ℎ𝑠 , and 𝑡 𝑡 ℎ𝑒

are also parameters that need to be optimized. The payload mass is already determined for a specific mission, so the

carrying performance is taken as a terminal constraint, not as the objective.

The above planning problem has segmented dynamics and complex constraints, and the influence of aerodynamic

forces and the landing process of debris are difficult to be analytically or parameterized expressed. Thus, the problem is

generally solved by numerical and iterative computations.

IV. Onboard Load Relief Control

The wind varies in space and time, so its profile is broken down into steady-state (mean) and unsteady components

to simplify the analysis [19]. The blue lines in Fig. 5 represent the wind speed and direction at different altitudes, and

the black lines represent the steady wind. The AOA generated by the steady wind represented by the black dotted lines

could be neglected (close to 0) after the wind biasing technique is applied. However, there are differences between the

black and blue curves, indicating that unsteady components exist, and the steady wind may also alter before liftoff, these

are exactly what the onboard load relief needs to address.

The active load relief is realized through attitude control, as shown in Fig. 6. The wind biasing trajectory provides

the reference, and the tracking method is applied inflight in the atmosphere to reduce the total AOA. Two methods are

applied for real-time load relief. One is LRA, widely used in the China’s new generation launchers, such as LM-7; the

other is LRE. When the wind acts on a rocket, it causes the rocket to drift and rotate, then the extended state observer

(ESO) provides a fairly general algorithm to estimate the disturbance torques other than the control torques if the model

is known. This solution originates from anti-disturbance rejection control (ADRC) technology [40], and in the strong

wind area and transonic regime, these disturbance torques can all be regarded as being caused by aerodynamic forces.

Thus, the LRA and LRE deal with the disturbance forces and moments caused by wind, respectively, which is similar to

13
 24 24
实测风
Actual
Actualwind
wind
Compensate wind
Compensated wind 实测风
Actual
Actualwind
wind
弹道修正风 Compensated
Compensatewind
wind
21 21 弹道修正风

18 18

15 15
（km）
(km)
高度（km）

高度（km）
Altitude (km)
(km)
Altitude

12 12
wind direction
Altitude

Altitude
shear
9 9
风速切变
wind speed shear
6 6

3 3

0 0
0 5 10 15 20 25 30 35 0 45 90 135 180 225 270 315 360
风速（m/s） 风向（°
wind ） (°)
direction
wind
Wind speed
speed(m/s)
(m/s) Wind direction (deg)

Fig. 5 Relationship between load relief and wind compensation.

the intention of the Ares I AD algorithm [27].

Wind biasing trajectory Onboard computer and embedded

attitude control
Tracking PD controller
reference - Transfer to body
software
Inertial Gain
measurement + coordinate system + Corrective
unit Attitude + network
Gain
Gyros

+
Amplitude Corrective Command
Accelerators Filter Gain
limiting network + + assign
Load relief based on lateral acceleration

Rate gyros ESO for Control Amplitude

acceleration law limiting
Load relief based on ESO
Onboard load relief

Dynamics Servo deflection

Fig. 6 Relationship between load relief and wind compensation.

A. Load Relief Based on Acceleration (LRA) and Load Relief Based on ESO (LRE)

• Principle of LRA

To simplify the discussion, the structural interference, aerodynamic damping, pitching moment caused by elastic

and propellant sloshing motions, and other uncertainties are ignored. Thus, the attitude dynamics equations in the pitch

14
channel are as follows:
 0𝜙
Δ𝜃¤ = 𝑐 1 Δ𝛼 + 𝑐 2 Δ𝜃 + 𝑐 3 𝛿 𝜙 + 𝑐 1 𝛼 𝑤

 𝜙 𝜙 𝜙







Δ𝜙¥ + 𝑏 1 Δ𝜙¤ + 𝑏 2 Δ𝛼 + 𝑏 3 𝛿 𝜙 = −𝑏 2 𝛼 𝑤 ,
𝜙 𝜙 𝜙 𝜙 (18)







 Δ𝜙 = Δ𝛼 + Δ𝜃

where Δ𝜃, Δ𝛼, Δ𝜙, and 𝛿 𝜙 represent the deviation of trajectory tilt angle, the deviation of AOA caused by attitude motion,
𝜙 𝜙 0𝜙
the deviation of pitch angle, and the engine swing angle, respectively; 𝑐 1 , 𝑐 2 , 𝑐 1 are the aerodynamic coefficients
𝜙 𝜙 𝜙 𝜙
related to AOA, ballistic inclination, and wind AOA, respectively; 𝑐 3 denotes the control force coefficient; 𝑏 1 , 𝑏 2 , 𝑏 3

are the coefficients of damping torque, aerodynamic torque, and control torque, respectively. Lateral acceleration, 𝑦¥ 1𝑎 ,

is measured by the accelerometers of the inertial measurement unit installed in the instrument bay, shown as follows:

𝜙 𝜙 ¥
𝑦¥ 1𝑎 = 𝑘 2 (Δ𝛼 + 𝛼 𝑤 ) + 𝑘 3 𝛿 𝜙 + 𝑙 𝑎𝑥 Δ𝜙, (19)

𝜙 𝜙
where 𝑙 𝑎𝑥 is the distance between inertial measurement unit and mass center, 𝑘 2 , 𝑘 3 represent the coefficients for

acceleration measurement related to AOA and control force, respectively.

The pitch control equation with LRA is as follows:

𝛿 𝜙 = 𝑎 0 Δ𝜙 + 𝑎 1 Δ𝜙¤ + 𝑎 𝑔 𝑦¥ 1𝑎 ,
𝜙 𝜙 𝜙
(20)

𝜙 𝜙 𝜙
where 𝑎 0 , 𝑎 1 are the static and dynamic gains, respectively; 𝑎 𝑔 represents the gain of LRA.

• Principle of LRE
¤ and 𝑏 is neglected (usually it is a small quantity). The
Based on the principle of an ESO [40], 𝑥1 = Δ𝜙, 𝑥2 = Δ𝜙,
𝜙
1

following equation can be obtained from Eq. (18):

𝑥¤2 = Δ𝜙¥ = −𝑏 2 (Δ𝛼 + 𝛼 𝑤 ) − 𝑏 3 𝛿 𝜙 .

𝜙 𝜙
(21)

𝜙 𝜙
With 𝑥 3 = −𝑏 2 (Δ𝛼 + 𝛼 𝑤 ), 𝑏 = −𝑏 3 and 𝑢 = 𝛿 𝜙 , Eq. (21) can be rewritten as follows:



𝑥¤1 = 𝑥2








 𝑥¤2 = 𝑥3 + 𝑏𝑢



, (22)




 𝑥¤3 = 𝑀0 (𝑡)




 𝑦 = 𝑥1




15
where 𝑀0 (𝑡) is unknown. Thus, the ESO is designed as follows:



𝑒 = 𝑧1 − Δ𝜙








𝑧¤1 = 𝑧 2 − 𝑙 0 𝑒




, (23)




 𝑧¤2 = 𝑧3 − 𝑙 1 𝑒 + 𝑏𝑢






 𝑧¤3 = −𝑙 2 𝑒


where 𝑙 0 , 𝑙1 and 𝑙2 are the observer parameters.

The parameters in Eq. (23) can be nonlinear, but linear parameters are used in this paper. In the ESO design, linear

ADRC technology is often regarded as an improved proportional–integral–derivative control, and detailed discussions

of this can be found elsewhere [41, 42]. If 𝑙 0 , 𝑙 1 and 𝑙 2 are designed appropriately, it can ensure that 𝑒 → 0, and then
¤ and 𝑧3 → −𝑏 𝜙 (Δ𝛼 + 𝛼 𝑤 ). Thus, the pitch control equation with the LRE is as follows:
𝑧1 → Δ𝜙, 𝑧 2 → Δ𝜙, 2

𝛿 𝜙 = 𝑎 0 Δ𝜙 + 𝑎 1 Δ𝜙¤ + 𝑎 3 𝑧 3 /𝑏 3 ,
𝜙 𝜙 𝜙
(24)

where 𝑎 3 is the gain of LRE.

B. Load Relief Analysis Under Steady and Shear Wind

• Steady wind

The steady wind is regarded as a slowly varying disturbance, so the dynamic terms of the rocket’s motion and control

equations can be ignored:

Δ𝜃¤ ≈ 0, Δ𝜙¤ ≈ 0, Δ𝜙¥ ≈ 0. (25)

Thus, Eq. (18) can be rewritten as follows:


 0𝜙
𝜙 𝜙 𝜙
𝑐 1 Δ𝛼 + 𝑐 2 Δ𝜃 + 𝑐 3 𝛿 𝜙 + 𝑐 1 𝛼 𝑤 𝑝 = 0








𝜙 𝜙 𝜙
𝑏 2 Δ𝛼 + 𝑏 3 𝛿 𝜙 = −𝑏 2 𝛼 𝑤 𝑝 , (26)







 Δ𝜙 = Δ𝛼 + Δ𝜃


where 𝛼 𝑤 𝑝 denotes the AOA caused by steady wind.

The feedback with the LRA is as follows:

𝜙 𝜙 𝜙 𝜙
𝛿 𝜙 = 𝑎 0 Δ𝜙 + 𝑎 𝑔 [𝑘 2 (Δ𝛼 + 𝛼 𝑤 𝑝 ) + 𝑘 3 𝛿 𝜙 ]. (27)

16
According to Eqs. (26) and (27), the total AOA under a steady wind can be obtained as follows:

𝜙 𝜙 𝜙 𝜙 𝜙 0𝜙 𝜙
𝑎 0 𝑏 3 (𝑐 1 − 𝑐 2 ) − 𝑎 0 𝑐 1 𝑏 3
𝛼ˆ 𝑎 𝑝 = 𝛼 𝑤 𝑝 + Δ𝛼 = 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙
𝛼𝑤 𝑝 = 𝑓𝑎 𝑝 𝛼𝑤 𝑝 , (28)
𝑎 0 𝑏 3 (𝑐 1 − 𝑐 2 ) − 𝑏 2 (𝑎 0 𝑐 3 + 𝑐 2 ) − 𝑎 𝑔 𝑐 2 (𝑏 3 𝑘 2 − 𝑏 2 𝑘 3 )

where 𝛼ˆ 𝑎 𝑝 , 𝑓 𝑎 𝑝 represent the total AOA and regulatory factor with LRA under steady wind, respectively.

Similarly, the feedback with the LRE is

𝜙 𝜙 𝜙
𝛿 𝜙 = 𝑎 0 Δ𝜙 − 𝑎 3 𝑏 2 (Δ𝛼 + 𝛼 𝑤 𝑝 )/𝑏 3 . (29)

From Eqs. (26) and (29), we obtain the following:

𝜙 𝜙 𝜙 𝜙 0𝜙
𝑎 0 𝑏 3 (𝑐 1 − 𝑐 2 − 𝑐 1 )
𝛼ˆ 𝜙¥ 𝑝 = 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙
𝛼 𝑤 𝑝 = 𝑓 𝜙¥ 𝑝 𝛼 𝑤 𝑝 , (30)
𝑎 0 𝑏 3 (𝑐 1 − 𝑐 2 ) − 𝑏 2 (𝑎 0 𝑐 3 + 𝑐 2 ) + 𝑎 3 𝑏 2 𝑐 2

where 𝛼ˆ 𝜙¥ 𝑝 , 𝑓 𝜙¥ 𝑝 represent the total AOA and regulatory factor with LRE under steady wind, respectively.

However, the above conclusions are derived under the assumptions given by Eq. (25). During a real flight, due to the

existence of variable aerodynamic forces and engine swing torque, these assumptions might not hold in general. Thus,

Eqs. (28) and (30) can only be used as a reference for the ideal state.

Considering that the steady wind is mainly compensated by the wind biasing trajectory, so only the shear wind is

discussed below.

• Shear wind

The motion of the center of mass is neglected because it changes slower than the shear wind, and Eq. (18) is

simplified as follows:

 𝜙 𝜙 𝜙
 𝑏 2 Δ𝛼 + 𝑏 3 𝛿 𝜙 = −𝑏 2 𝛼 𝑤𝑞



, (31)



 Δ𝜙 = Δ𝛼

where 𝛼 𝑤 𝑞 denotes the AOA caused by shear wind.

Based on Eqs. (31) and (27), the AOA with the LRA under shear wind is deduced:

𝜙 𝜙
𝑎0 𝑏3
𝛼ˆ 𝑎𝑞 = 𝛼 𝑤 𝑞 + Δ𝛼 = 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙
𝛼 𝑤𝑞 = 𝑓 𝑎𝑞 𝛼 𝑤𝑞 , (32)
𝑎 0 𝑏 3 + 𝑏 2 + 𝑎 𝑔 (𝑏 3 𝑘 2 − 𝑏 2 𝑘 3 )

where 𝛼ˆ 𝑎𝑞 , 𝑓 𝑎𝑞 represent the total AOA and regulatory factor with LRA under the shear wind, respectively.

Based on Eqs. (31) and (29), the AOA with the LRE under shear wind is as follows:

𝜙 𝜙
𝑎0 𝑏3
𝛼ˆ 𝜙𝑞
¥ = 𝜙 𝜙 𝜙 𝜙
𝛼 𝑤𝑞 = 𝑓 𝜙𝑞
¥ 𝛼 𝑤𝑞 , (33)
𝑎0 𝑏3 + 𝑏2 − 𝑎3 𝑏2

17
where 𝛼ˆ 𝜙𝑞
¥ , 𝑓 𝜙𝑞
¥ represent the total AOA and regulatory factor with LRE under the shear wind, respectively.
𝜙 𝜙 𝜙 𝜙
For most launchers, 𝑏 3 𝑘 2 − 𝑏 2 𝑘 3 > 0. Several features of Eqs. (32) and (33) should be highlighted.
𝜙 𝜙
Remark 1. The load relief effect of the LRA on the shear wind is related to the vehicle’s parameters 𝑘 𝑖 , 𝑏 𝑖 (𝑖 = 2, 3)
𝜙 𝜙
and the control parameter 𝑎 0 . The larger the feedback gain (𝑎 𝑔 > 0), the better the load relief effect.
𝜙
Remark 2. The load relief effect of the LRE on the shear wind is related to the vehicle’s parameters 𝑏 𝑖 (𝑖 = 2, 3)
𝜙 𝜙
and the control parameter 𝑎 0 . For an aerodynamically unstable rocket, 𝑏 2 < 0. Thus, the larger the feedback gain

(𝑎 3 > 0), the better the load relief effect.

𝜙
For an aerodynamically stable rocket (𝑏 2 > 0), the effect of the load relief depends on the absolute value of 𝑓 𝜙𝑞
¥ . If

both methods are adopted, the control equation of the pitch plane is as follows:

𝛿 𝜙 = 𝑎 0 Δ𝜙 + 𝑎 1 Δ𝜙¤ + 𝑎 3 𝑧3 /𝑏 3 + 𝑎 𝑔 𝑦¥ 1𝑎 .
𝜙 𝜙 𝜙 𝜙
(34)

The total AOA under shear wind is then

𝜙 𝜙
𝑎0 𝑏3
𝛼ˆ 0 = 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙 𝜙
𝛼 𝑤𝑞 = 𝑓𝑞 𝛼 𝑤𝑞 (35)
𝑎 0 𝑏 3 + 𝑏 2 + 𝑎 𝑔 (𝑏 3 𝑘 2 − 𝑏 2 𝑘 3 ) − 𝑎 3 𝑏 2

where 𝛼ˆ 0 , 𝑓𝑞 represent the total AOA and regulatory factor with LRA/LRE under shear wind, respectively.

Remark 3. The influence of the combination of the two methods under the shear wind is considered. Generally,
𝜙 𝜙
𝑎 𝑔 > 0 and 𝑎 3 > 0. For an aerodynamically unstable rocket, 𝑏 2 < 0, and thus, 0 < 𝑓𝑞 < 𝑓 𝑎𝑞 and 0 < 𝑓𝑞 < 𝑓 𝜙𝑞
¥ , which

means the load relief effect is better than that of either method applied alone. For an aerodynamically stable rocket,
𝜙
𝑏 2 > 0, and the effect of the combined load relief methods depends on the absolute value of 𝑓𝑞 .

C. Selection of Load Relief Gain in LRE

Eq. (23) can also be re-formulated as follows:



𝑒 = 𝜔 𝑧1 − 𝑧 2








𝑧¤2 = 𝑧 3 + 𝑙1 𝑒 + 𝑏𝑢 , (36)







 𝑧¤3 = 𝑙 2 𝑒


where 𝜔 𝑧1 denotes the pitch angular velocity. When 𝑒 → 0, 𝑧2 → 𝜔 𝑧1 ≈ Δ𝜙¤ (assuming that the rotational motion is

much faster than the translational motion).

It should be noted that when considering the stable effect of the load relief, it is approximately regarded as
𝜙
𝑧3 ≈ −𝑏 2 (Δ𝛼 + 𝛼 𝑤 ). However, in this section, this assumption does not hold because the transit state cannot be ignored

for stability analysis.

18
 • Considering impact on stability

The following Laplace transforms are obtained from Eq. (36):



𝜔 𝑧1 = 𝑠Δ𝜙








𝑠𝑧 2 = 𝑧3 + 𝑙 1 (𝑠Δ𝜙 − 𝑧2 ) − 𝑏 3 𝛿 𝜙 . (37)







 𝑠𝑧 3 = 𝑙2 (𝑠Δ𝜙 − 𝑧 2 )


Based on Eqs. (18), (24), and (37), the characteristic equation of the closed-loop system, or the denominator of

closed-loop transfer function ΔΦ(𝑠)/𝛿Φ (𝑠), is as follows:

𝐷 (𝑠) = 𝐷 4 𝑠4 + 𝐷 3 𝑠3 + 𝐷 2 𝑠2 + 𝐷 1 𝑠 + 𝐷 0 , (38)

where ΔΦ(𝑠) and 𝛿Φ (𝑠) are the Laplace transformation of Δ𝜙 and 𝛿 𝜙 , respectively. The superscripts of the variables are

omitted, and 𝐷 0 = 𝑙 2 𝑏 2 (1 − 𝑎 3 ) +𝑙2 𝑏 3 𝑎 0 , 𝐷 1 = 𝑙2 𝑏 1 (1 − 𝑎 3 ) +𝑙 2 𝑏 3 𝑎 1 +𝑙 1 𝑏 2 +𝑙 1 𝑏 3 𝑎 0 , 𝐷 2 = 𝑙 2 +𝑙1 𝑏 1 +𝑏 2 +𝑙1 𝑏 3 𝑎 1 +𝑏 3 𝑎 0 ,

𝐷 3 = 𝑙 1 + 𝑏 1 + 𝑏 3 𝑎 1 , and 𝐷 4 = 1. According to the Hurwitz stability criterion, the condition that the root of 𝐷 (𝑠) = 0

has no negative real part in the complex plane is as follows:



𝐷 0 > 0, 𝐷 1 > 0, 𝐷 2 > 0, 𝐷 3 > 0








𝐷𝐷 1 = 𝐷 2 𝐷 3 − 𝐷 1 > 0 . (39)




 𝐷𝐷 2 = 𝐷 1 𝐷 2 𝐷 3 − 𝐷 21 − 𝐷 0 𝐷 23 > 0





For a stable system, 𝑎 0 , 𝑎 1 , 𝑙 1 , and 𝑙 2 are all greater than 0. 𝑏 1 and 𝑏 3 are also greater than 0. For an aerodynamically

stable vehicle, 𝑏 2 > 0; for an aerodynamically unstable vehicle, 𝑏 2 < 0. Taking the latter as an example, the derivatives

of the above coefficients at the time of maximum dynamic pressure are as follows:


 𝜕𝐷 𝜕𝐷1 𝜕𝐷2 𝜕𝐷3
 𝑎3 0 = −𝑙 2 𝑏 2 > 0, = −𝑙 2 𝑏 1 < 0, = = 0;



 𝑎3 𝑎3 𝑎3



𝜕(𝐷𝐷1 ) (40)
 𝑎3 = − 𝜕𝐷 1
𝑎3 > 0
.



𝜕(𝐷𝐷2 )
= −𝑙 2 𝑏 1 𝐷 2 𝐷 3 + 2𝑙2 𝑏 1 𝐷 1 + 𝑙2 𝑏 2 𝐷 23



 𝑎3


Since 𝜕𝐷 1 /𝑎 3 < 0, with the increase in 𝑎 3 , 𝐷 1 would be less than 0, which would cause system instability.

• Considering influence on attitude control accuracy

If the influence of 𝑏 1 is ignored, the closed-loop transfer function from the aerodynamic AOA to the attitude

19
deviation is established as follows:

−𝑏 2 𝑠2 + 𝑙1 𝑠 + 𝑙2 (1 − 𝑎 3 )


ΔΦ (𝑠) −𝑏 2 𝑠2 − 𝑏 2 𝑙1 𝑠 − 𝑏 2 𝑙2 (1 − 𝑎 3 )
= 0 4 = 2
, (41)
𝐷 4 𝑠 + 𝐷 03 𝑠3 + 𝐷 02 𝑠2 + 𝐷 01 𝑠 + 𝐷 00


𝐴 (𝑠) 𝑠 + 𝑙1 𝑠 + 𝑙2 𝑠2 + 𝑏 3 𝑎 1 𝑠 + 𝑏 3 𝑎 0

where 𝐴 (𝑠) is the Laplace transformation of total angle of attack , and 𝐷 00 = 𝑙2 𝑏 3 𝑎 0 , 𝐷 01 = 𝑙 2 𝑏 3 𝑎 1 + 𝑙 1 𝑏 3 𝑎 0 ,

𝐷 02 = 𝑙2 + 𝑙 1 𝑏 3 𝑎 1 + 𝑏 3 𝑎 0 , 𝐷 03 = 𝑙1 + 𝑏 3 𝑎 1 , and 𝐷 04 = 1.

If 𝑎 3 = 0, it is equivalent to the case without the LRE. Thus, 𝐸 (𝑠) is a function included in the transfer function

when the ESO feedback is introduced:

𝑠2 + 𝑙 1 𝑠 + 𝑙 2 (1 − 𝑎 3 )
𝐸 (𝑠) = . (42)
𝑠2 + 𝑙1 𝑠 + 𝑙2

Under the condition that 𝑙 1 and 𝑙2 are selected and 𝑙1 , 𝑙2 > 0, the following can be determined and 𝐴𝑒 denotes the

amplitude value:

1) if 0 < 𝑎 3 < 2, then 𝐴𝑒 (𝐸 (𝑠)) < 1, and it is conducive to the stability of rigid body attitude motion;

2) if 𝑎 3 ≥ 2, then 𝐴𝑒 (𝐸 (𝑠)) ≥ 1, and the attitude angle deviation would increase;

3) if 𝑎 3 = 1, the attitude deviation would be the smallest.

V. Simulations and Flight Results Analysis

A. Load Relief Effect Simulation

According to the statistical wind field of the Wenchang launch site, the effects of various load relief methods for the

LM-8 were evaluated by simulations and are compared in Table 1.

Table 1 Comparison of load relief effects (normalized)

Max 𝑞𝛼 in simulations Load relief

No.
No throttling Throttling methods
1 1 93.1% WB LRA
2 82.1% 70.5% WB LRA
3 86.2% 74.8% WB LRA
4 68.4% 57.3% WB LRA
Note: WB denotes wind biasing. A black square indicates that the method was applied.

For the LRE, all the resultant moments except the control moment were identified by 𝑧 3 , including the pitch moment

generated by elastic and sloshing motions, which need to be filtered to obtain the swing angle generated by the total

AOA:
𝜙 𝜙
𝛿 𝜙 𝐴𝐷𝑅𝐶 = 𝐺 (𝑠) (𝑧3 + 𝑏 1 𝑧2 )/𝑏 3 , (43)

𝜙
where 𝐺 (𝑠) is the filter network, and 𝑏 1 𝑧2 is the damping torque. Thus, the ideal feedback of Eq. (24) can also be

20
written as follows in practice:

𝛿 𝜙 = 𝑎 0 Δ𝜙 + 𝑎 1 Δ𝜙¤ + 𝑎 3 𝛿 𝜙 𝐴𝐷𝑅𝐶 .
𝜙 𝜙
(44)

Based on the bandwidth of the observer 𝜔 𝐸𝑆𝑂 , we set 𝑙 0 = 3𝜔 𝐸𝑆𝑂 , 𝑙1 = 3𝜔2𝐸𝑆𝑂 , and 𝑙2 = 𝜔3𝐸𝑆𝑂 .

Table 2 shows the summary of the simulation results under the LRE and the LRA, in which 𝑆𝑡𝑟𝑢 𝑃 , 𝑆𝑡𝑟𝑢 𝑁 , and

𝑆𝑡𝑟𝑢 0 represent the positive, negative, and none of the structural interference, respectively. The coefficient 𝑏 2 is the

distance from the pressure center to the center of mass. The data in Table 2 are the percentage decreases in the 𝑞𝛼 value

after adopting the load relief method compared to that without the load alleviation solutions.

Table 2 Load relief effect comparison

Aerodynamic LRE LRA LRE + LRA

No.
instability coefficient Stru 𝑃 Stru0 Stru 𝑁 Stru 𝑃 Stru0 Stru 𝑁 Stru 𝑃 Stru 𝑁
1 1.5𝑏 2 23.42% 23.51% 23.39% 23.19% 23.59% 23.78% 39.41% 39.92%
2 𝒃2 11.21% 10.87% 10.47% 15.74% 15.99% 16.21% 25.13% 25.23%
3 0 1.6% −1.09% 0.52% 9.20% 8.41% 9.48% 10.63% 10.35%
4 -𝑏 2 4.73% −5.03% −5.10% 10.22% 6.04% 6.21% 5.95% 1.70%

When 𝑏 2 was increased (e.g., the data in the first row of Table 2), the effect of the LRE was significantly improved.

In contrast, when 𝑏 2 was decreased (e.g., the data in third row of Table 2), or when the aerodynamic instability changed

to stability (i.e., the polarity of 𝑏 2 was adjusted, as shown in the fourth row), the load reduction effect decreased (under

the assumption that the feedback gains remained the same), and in some situations the 𝑞𝛼 increased (see the negative

value in Table 2). These results are consistent with Remarks 1 and 2.

For the combined LRE and LRA method, the combined load reduction effect was better than that of either single

method. However, in the case of aerodynamic stability (i.e., for 𝑏 2 > 0), if 𝑎 3 > 0, the combined effect is worse than

that of the LRA. The result is consistent with Remarks 3. It can also be concluded that when the aerodynamic force

was much greater than the structural interference, the influence of the latter can be almost ignored.

The second line in Table 2 shows the nominal scenario of LM-8. The effect of the LRA was better than that of

the LRE. However, this conclusion was not consistent with their load relief factors, which are compared in Fig. 7 (a).

During the load relief period, 𝑓 𝑎𝑞 > 𝑓 𝜙𝑞

¥ , which means that the LRE was better, contradicting the results in Table

2. The reason for this discrepancy lies in the filter design of the ESO processing, which was expected to filter the

moments generated by elasticity and propellant sloshing so that only the aerodynamic force moment remained. However,

constrained by the bandwidth of the filter, the effect was degraded, and thus, there was no close relationship between the

𝑞𝛼 values and the variations of 𝑓 𝑎𝑞 and 𝑓 𝜙𝑞

¥ , as shown in Fig. 7 (b).

To further analyze the data in Fig. 7, the theoretical moments (angular acceleration) caused by aerodynamic forces

in the simulation were extracted and compared with the moments predicted by the ESO, as shown in Fig. 8 (a). There

were deviations in the amplitudes and phases of the ESO filtering result and the theoretical values. Fig. 8 (b) can be

21
 2
f_acc + f_ESO f q -f_ESO
fq
fq q.α_acc - q.α_ESO f_acc 1.3
1.8 f_acc 160
f q
f_ESO
fq
1.6 1.1
100
Regulatory factor (/)

1.4 0.9
40
1.2
0.7

Pa.rad
1 -20
0.5
0.8
-80
0.3
0.6

-140 0.1
0.4

0.2
20 30 40 50 60 70 80 90 100 110 120 -200 -0.1
0 20 40 60 80 100 120
Time(s)
Time(s)

(a) Comparison of regulator factors under shear wind (b) Comparison between 𝑞 𝛼 and load relief factors

Fig. 7 Comparison of the regulatory factors and their load relief effects.

regarded as the load relief results of two types of data in Fig. 8 (a). As shown in Fig. 8 (b), the values of 𝑞𝛼 represented

by the blue line were smaller than those represented by the red line, which means the load relief based on the theoretical

moments achieved a total reduction of 20.54%, even better than that of the LRA (the second line in Table 2).

4 1600
  dd
model_
theo
q theo
qa_model
ESO
1400 q ESO
deg/s 2 of pitch (deg/s )

 dd qa_ESO
2

ESO_
3
1200

1000
2
800
Pa.rad
Angular acceleration

1 600

400
0
200

0
-1
-200

-2 -400
40 50 60 70 80 90 100 10 20 30 40 50 60 70 80
Time (s)
Times(s) Time
Time(s)(s)

(a) Comparison between theoretical and estimated angular acceleration (b) 𝑞 𝛼 based on theoretical and estimated angular acceleration

Fig. 8 Comparison of the estimation errors and 𝑞𝛼 values.

Based on the above analysis, the following conclusions can be drawn:

1) Under ideal conditions, the LRE had a better effect than the LRA (as shown in Fig. 7 (a)). However, when

considering the dynamic process and ESO filtering, its effect was degraded. The effect of the filtering network

was affected by the frequency interception.

2) Tracking lag was found in the ESO estimations of the angular accelerations (as shown in Fig. 8 (a)). If the

aerodynamic moment can be determined more accurately, the load relief effect can be further improved (as

shown in Fig. 8 (b)).

22
 At present, the lag caused the filter to detect a smaller rising signal and a larger falling signal than the corresponding

theoretical values. Some innovations on the observer design are needed to make more accurate estimations in the future.

B. Influence of Extended State Observer (ESO) on Stability and Control Accuracy

With 𝑙1 = 1.6 and 𝑙 2 = 0.6, the influence of the ESO on the stability is studied here, and variations of the above

coefficients of Eq. (39) with the increase in 𝑎 3 are shown in Fig. 9. 𝐷 1 and 𝐷𝐷 2 gradually became less than 0, leading

to attitude control instability. Thus, the upper limits of 𝑎 3 could be determined.

2 4 6

3
1 5
2
D0

D1

D2
0 4
1

-1 0 3
0 50 0 50 0 50
a3 a3 a3

4 11 20

3.5 15
10
3 10
DD1

DD2
D3

2.5 5
9
2 0

1.5 8 -5
0 50 0 50 0 50
a3 a3 a3

Fig. 9 Variations of coefficients.

Similarly, the influence of 𝑎 3 on the control accuracy at the moment of the maximum dynamic pressure was

considered and are shown in Fig. 10, with 𝑎 3 varying from 0.5 to 1.5. The attitude motion was stable after ESO feedback

was adopted. In the low-frequency regions, the amplitude of the attitude angle deviation caused by the AOA decreased

when 𝑎 3 changed from 0.5 to 1 and increased when it changed from 1 to 1.5. With the increase in the frequency, the

attenuation of the attitude angle deviation was weakened gradually.

C. Flight Results

For LM-8, the deeper the engine throttling ability was, the more effectively it could reduce the maximum dynamic

pressure. However, considering the influence on the working stability of the engine, a widely accepted throttling level,

i.e., 77.5%, is scheduled for YF100 engines. In addition, the earlier the throttling time is, the more effectively it reduces

the maximum dynamic pressure, but the greater the loss of capacity. Thus, after considering various trade-offs, the final

23
 5
Frequencies increase
0

-5

-10

Magnitude(dB) -15

-20

-25

-30

-35
0 20 40 60 80 100 120 140 160 180
Phase(deg)

Fig. 10 Nichols diagram of 𝐸 (𝑠).

throttling time was set to 35-135 s.

The flow regulator had sufficient thrust adjustment accuracy. With the ground test data as an example, the measured

thrust value throttled to 77.5% was 936.7 kN (the reference baseline value was 1214.8 kN), with an error of 0.5%.

When it was throttled to 75% and 80%, the thrust was 903.7 and 958.9 kN (the reference baseline value was 1206.1 kN),

respectively, with corresponding errors of 0.1% and 0.6%.

According to the flight telemetry data and the measured wind field 30 min after launch, the load relief effect was

reproduced, as shown in Table 3. The maximum 𝑞𝛼 on launch day was reduced to 1183 Pa·rad, which ensured the

successful debut of LM-8.

Table 3 Load relief effect (normalized) of the LM-8 debut*

39 s (wind biasing 52 s (shear on 58 s (shear on 75 s (shear on Load relief

No.
introduced) wind direction) wind speed) wind direction) methods
1 93.3% 82.8% 93.8% 75% WB LRA LRE
2 88.2% 68.6% 90% 63% WB LRA LRE
3 78.4% 57.1% 88% 54.9% WB LRA LRE
* All data are compared to the maximum value of 𝑞𝛼, which is simulated with no load relief measures applied at the
corresponding moment.

VI. Conclusions
This paper introduces a comprehensive load relief scheme to meet the flight load requirements of a legacy stage

under a new flight profile. A unified processing procedure suitable for gust and fluctuating pressure load analysis

24
is introduced. Various models, such as engine throttling and wind biasing models, were integrated into an integral

trajectory planning problem to simultaneously optimize 𝑞𝛼 while ensuring the launching performance of the prescribed

payload and improving the adaptability to various uncertainties.

The LRE’s principle, effect, and influence on the stability were thoroughly studied. For a rocket with large

aerodynamic instability coefficient, such as LM-8, the greater the LRE feedback gain, the better the load relief effect.

The effect was also closely related to the rocket parameters and control law gain, which were time-varying during flight,

requiring the load relief feedback gain to be adjusted accordingly. It can also be set as a constant based on the maximum

𝑞𝛼 scenario while ensuring that 𝑞𝛼 does not exceed the limit in other time periods. However, the continuous increase in

the feedback gain will affect the control accuracy and the attitude control stability.

To clearly reveal the principle of load relief methods, the dynamics of the launcher is simplified, and the features

of the structural flexibility and liquid sloshing, whose effects on the load alleviation could be studied in the future,

are ignored. However, this simplification is acceptable, because the load relief regulation is relatively slower than the

elastic and sloshing motions, and all the evaluation results are based on the models including the elastic and sloshing

factors. The successful application of the above measures not only provides convenience for the modular re-assembly of

the launch vehicle, but also provides a systematic solution for in-service rockets to further improve the environmental

adaptability and increase the launch probability.

Acknowledgements
This work is supported by the China Aerospace Science and Technology Corporation in relation to the development

plan of LM-8.

References
[1] Wang, X. J., Fan, R. X., Cheng, T. M., Xu, L. J., and Yang, H. L., “Maiden flight of Long March 7-The new generation medium

launch vehicle in China,” Aerospace China, Vol. 17, No. 3, 2016, pp. 12–18. URL http://www.aerospacechina.org/CN/

Y2016/V17/I3/12.

[2] Li, D., Li, P. Q., Wang, J., Huang, B., and Liu, B., “General scheme and key technology of Long March 5 launch vehicle

(in Chinese),” Journal of Deep Space Exploration, Vol. 8, No. 4, 2021, pp. 335–343. https://doi.org/10.15982/j.issn.2096-

9287.2021.20210009.

[3] Song, Z. Y., Wu, Y. T., Xu, S. S., Chen, X. F., and Y, X., “LM-8: the Pioneer of Long March rocket series on the innovations

of Commercialization and intelligence (in Chinese),” Journal of Deep Space Exploration, Vol. 8, No. 1, 2021, pp. 3–16.

https://doi.org/10.15982/j.issn.2096-9287.2021.20200009.

[4] Vause, R. F., and Brett, R. S., “Trajectory-Based Loads for the Ares I-X Test Flight Vehicle,” AIAA Atmospheric Flight

Mechanics Conference, Portland,OR, Auguste 8-11, 2011. https://doi.org/10.2514/6.2011-6463.

25
 [5] Brett, R. S., Isam, Y., and Aaron, D. O., “Use of Flexible Body Coupled Loads in Assessment of Day of Launch Flight Loads,”

AIAA Atmospheric Flight Mechanics Conference, Portland,OR, Auguste 8-11, 2011. https://doi.org/10.2514/6.2011-6465.

[6] Decker, R., Duffin, P., Hill, A., Beck, R., and Dukeman, G. (eds.), SLS Trade Study 0058: Day of Launch (DOL) Wind Biasing,

M14-3408, June 30, 2014. URL https://ntrs.nasa.gov/citations/20140008595.

[7] Decker, R., Leahy, F., Dukeman, G., Hill, A., and Beck, R. (eds.), ESD and SLS Natural Environments Update & SLS

Day-of-Launch (DOL) Wind Biasing Launch Availability with Launch Window Effects, M14-4057, September 11, 2014. URL

https://ntrs.nasa.gov/citations/20140016513.

[8] Ryan, R. S., Mowery, D. K., and Winder, S. W. (eds.), Fundamental conncepts of structural loading and load relief techniques

for the space shuttle, NASA-TM-X-64684, August 8, 1972. URL https://ntrs.nasa.gov/citations/19720024231.

[9] Vieweg, R. R. (ed.), Load relief and gust alleviation control study final report, NASA-CR-123646, April, 1972. URL

https://ntrs.nasa.gov/citations/19720017815.

[10] NASA, “Rocket Engine Innovations Advance Clean Energy,” Spinoff 2011, February 1, 2012. URL https://ntrs.nasa.gov/

citations/20120001874.

[11] Betts, E. M., and Frederick, R. A., “A historical systems study of liquid rocket engine throttling capabilities,” 46th

AIAA/ASME/SAE/ASEE Joint Propulsion Conference and Exhibit, Nashville, TN, July 25-28, 2010. https://doi.org/10.2514/6.

2010-6541.

[12] Ernsberger, G. (ed.), Wind Biasing Techniques for Use in Obtaining Load Relief, NASA-TM-X-64604, June 14, 1971. URL

https://ntrs.nasa.gov/citations/19710020627.

[13] Winchell, W. M., “Launch Winds Operations for the Delta Launch Vehicle,” 30th Aerospace Sciences Meeting and Exhibit,

Reno, NV, January 6-9,1992. https://doi.org/10.2514/6.1992-717.

[14] Pat Blanchet, P., and Bartos, B., “An improved load relief wind model for the Delta launch vehicle,” AIAA 2001-0841. 39th

Aerospace Sciences Meeting and Exhibit, January 8-11,2001. https://doi.org/10.2514/6.2001-841.

[15] Dhuyvetter, H. J., and Lam, H. T., “Load relief for launch vehicles using an engine biasing technique,” Aerospace Technology

Conference and Exposition, 1989. https://doi.org/10.4271/892305.

[16] Smith, O. E., Adelfang, S. I., Batts, G. W., and Hill, C. K. (eds.), Wind models for the NSTS ascent trajectory biasing for wind

load alleviation, NASA TM-100375, August 1989. URL https://ntrs.nasa.gov/citations/19890019180.

[17] L, Y. M., “CZ-2E ballistic correction for high altitude wind (in Chinese),” Missiles and Space Vehicles, , No. 249, 2001, pp.

9–15.

[18] Carpentier, B., Haensler, P. E., Mazellier, B., and Espinosa-Ramos, A., “Loads alleviation on European launchers using wind

biased trajectory optimization,” 23rd AAS/AIAA Space flight Mechanics Meeting, 2013, pp. 321–336.

26
[19] Johnson, D. L., and Vaughan, W. W., “The wind Environment Interactions Relative to Launch Vehicle Design,” Journal of

Aerospace Technology Management, Vol. 12, 2020, p. e0220. https://doi.org/10.5028/jatm.v12.1090.

[20] Sharp, J. (ed.), Attitude Control Systems for Load Relief of Saturn-Class Launch Vehicles, NASA-CR-61375, February 1971.

URL https://ntrs.nasa.gov/citations/19720011005.

[21] Rheinfurth, M., “The Alleviation of Aerodynamic Loads on Rigid Space Vehicles,” Conference on Aerospace Meteorology, Los

Angeles, CA, 1966. https://doi.org/10.2514/6.1966-350.

[22] Song, Z. Y., “Load control technology in launch vehicle (in Chinese),” Aerospace Control, Vol. 31, No. 5, 2013, pp. 3–7.

[23] Suzuki, H., “Load Relief Control of H-IIA Launch Vehicle,” IFAC Proceedings Volumes, Vol. 37, No. 6, June 2004, pp.

985–990. https://doi.org/10.1016/S1474-6670(17)32307-8.

[24] Simplicio, P., Marcos, A., and Bennani, S., “Launcher flight control design using robust wind disturbance observation,” Acta

Astronautica, , No. 186, 2021, pp. 303–318. https://doi.org/10.1016/j.actaastro.2021.05.044.

[25] Jee, G., Brinda, V., Lalithambika, V. R., and Dhekane, M. V., “Achieving load relief with minimal change in conventional control

law of a launch vehicle,” IFAC-PapersOnLine, Vol. 49, No. 1, 2016, pp. 361–366. https://doi.org/10.1016/j.ifacol.2016.03.080.

[26] Oh, G., Park, J., Park, J., Lee, H., Kim, Y., Shin, S. J., Ahn, J., and Cho, S., “Load relief control of launch vehicle using

aerodynamic angle estimation,” Proc IMechE Part G: J Aerospace Engineering, Vol. 232, No. 8, 2018, pp. 1598–1605.

https://doi.org/10.1177/0954410017699435.

[27] Hall, C., Lee, C., Jackson, M., Whorton, M., and Compton, J., “Ares I Flight Control System Overview,” 2008 AIAA Guidance,

Navigation and Control Conference, Portland,OR, Auguste 18-21, 2008. https://doi.org/10.2514/6.2008-6287.

[28] Dennehy, N., and Orr, J. S., “Launch Vehicle Load Relief: A Historical Perspective and Some New Concepts,” NESC GN&C

Technical Discipline Team Webcast Series, Apr 02, 2021. URL https://ntrs.nasa.gov/citations/20210023589.

[29] Mastroddi, F., Stella, F., Cantiani, D., and Vetrano, F., “Linearized aeroelastic gust response analysis of a launch vehicle,”

Journal of Spacecraft and Rockets, Vol. 48, No. 3, 2011, pp. 420–432. https://doi.org/10.2514/1.47268.

[30] Kim, M. C., Kabe, A. M., and Lee, S. S., “Atmospheric Flight Gust Loads Analysis,” Journal of Spacecraft and Rockets, Vol. 37,

No. 4, July 2000, pp. 446–452. https://doi.org/10.2514/2.3603.

[31] Dynamic Gust Loads, AC No: 25.341-1, [DB/OL]. U.S. Department of Transportation Federal Aviation Administration,

Advisory Circular, 2014. URL https://www.faa.gov/documentLibrary/media/Advisory_Circular/AC_25_341-1.pdf.

[32] LIEPMANN, H. W., “On the Application of Statistical Concepts to the Buffeting Problem,” Journal of the Aeronautical

Sciences, Vol. 19, No. 12, 1952, pp. 793–800. https://doi.org/10.2514/8.2491.

[33] KÁRMÁN, V., and Lin, C., “On the statistical theory of isotropic turbulence,” Advances in Applied Mechanics, , No. 2, 1951,

pp. 1–19.

27
 [34] Piatak, D. J., Sekula, M. K., and Rausch, R. D., “Ares launch vehicle transonic buffet testing and analysis techniques,” 28th

AIAA Applied Aerodynamics Conference, Chicago, Illinois, 28 June–1 July, 2010. https://doi.org/10.2514/1.A32175.

[35] Liu, Y., Wang, G., Zhu, H. Y., and Ye, Z., “Numerical analysis of transonic buffet flow around a hammerhead payload fairing,”

Aerospace Science and Technology, Vol. 84, 2019, pp. 604–619. https://doi.org/10.1016/j.ast.2018.11.002.

[36] Piatak, D. J., and Sekula, M. K., “Measurement and analysis of terminal shock oscillation and buffet forcing functions on a launch

vehicle payload fairing,” AIAA Aviation 2019 Forum, Dallas, Texas, June 17-21, 2019. https://doi.org/10.2514/6.2019-3302.

[37] Timme, S., Badcock, K. J., and Da Ronch, A., “Gust analysis using computational fluid dynamics derived reduced order models,”

Journal of Fluids and Structures, Vol. 71, 2017, pp. 116–125. https://doi.org/10.1016/j.jfluidstructs.2017.03.004.

[38] Sako, B. H., Kabe, A. M., and Lee, S. S., “Statistical Combination of Time-Varying Loads,” AIAA Journal, Vol. 47, No. 10,

October 2009. https://doi.org/10.2514/1.39384.

[39] Li, B., Zhang, R., Zhang, M., Wang, H., Li, C., and Wang, K., “A review of throttling technology development for large-thrust

liquid rocket engines,” Aerospace China, Vol. 22, No. 2, 2021, pp. 16–26. https://doi.org/10.3969/j.issn.1671-0940.2021.02.003.

[40] Han, J. Q., Active disturbance rejection control technique - the technique for estimating and compensating the uncertainties (in

Chinese), Beijing: National Defense Industry Press, 2008.

[41] Jin, H. Y., Song, J. C., and Lan, W. Y., “On the characteristics of ADRC: a PID interpretation,” SCIENCE CHINA Information

Sciences, Vol. 63, 2020, p. 209201. https://doi.org/10.1007/s11432-018-9647-6.

[42] Zhong, S., Huang, Y., and Guo, L., “A parameter formula connecting PID and ADRC,” SCIENCE CHINA Information Sciences,

Vol. 63, 2020, p. 192203. https://doi.org/10.1007/s11432-019-2712-7.

28

View publication stats