Optimal tracking control for underactuated airship

A non-linear mathematical model of underactuated airship is derived in this paper based on Euler-Newton approach. The model is linearized with small disturbance theory, producing a linear time varying (LTV) model. The LTV model is verified by comparing its output response with the result of the nonlinear model for a given input signal. The verified LTV model is used in designing the LQT controller. The controller is designed to minimize the error between the output and required states response with acceptable control signals using a weighted cost function. Two LQT controllers are presented in this work based on two different costates transformations used in solving the differential Riccati equation (DRE). The first proposed assumption of costates transformation has a good tracking performance, but it is sensitive to the change of trajectory profile, whereas the second one overcomes this problem due to considering the trajectory dynamics. Therefore, the first assumption is performed across the whole trajectory tracking except for parts of trajectory profile changes where the second assumption is applied. The hybrid LQT controller is used and tested on circular, helical, and bowed trajectories. The simulation assured that the introduced hybrid controller results in improving airship performance.

Elfes et al. [9] showed that robotic airships had unique characteristics that make them ideal for exploring planets and moons with an atmosphere, allowing for precise flight path execution, long-range observations, transportation of scientific instruments, and opportunistic flight path replanning.Bueno et al. [10] provided an overview of Project AURORA, which focuses on the development of technologies for autonomous robotic airships for environmental monitoring and inspection applications.It covers various aspects such as hardware and software infrastructures, control and guidance methods, visual servoing strategies, and dynamic target recognition.Miura et al. [11] discussed the development of wireless access systems using high altitude platform stations for telecommunication and broadcasting purposes.Li et al. [12] discussed the use of unmanned airship image systems and processing techniques for identifying fresh water wetlands at a community scale.It highlights the limitations of satellite remote sensing images in extracting specific information about aquatic ecosystems and proposes the use of unmanned airship platforms with low-cost imaging instruments for higher spatial accuracy.Ilcev [13] examined the potential applications of stratospheric communication platforms (SCP) as an alternative for satellite communications, with various applications and services planned using aircraft or airship SCPs.SCPs can provide communication facilities for fixed and mobile applications, with commercial and military solutions.Subscribers can transmit and receive information through uplink to the platform, and the onboard SCP switching devices will route traffic to other subscribers within the same platform coverage or to other platforms or networks.SCPs can deploy antennas for large coverage areas or multibeam antennas for spot beams.Koska et al. [14] introduced the autonomous mapping airship equipped with lidar and compares it with other methods and technologies for mapping medium-sized areas.An overview of using high altitude platform as an international mobile telecommunication base station is introduced by Zhou et al. [15].
The model is linearized with small disturbance theory to design a suitable linear controller.The theory depends on the first derivative term of Taylor expansion.It assumes that the system behave linearly in a small portion of time so that the performance of the model can be estimated and optimized by a linear controller [28].Linear quadratic regulator (LQR) and LQT are optimal control techniques developed on the concept of calculus of variations to compute the best control gains to achieve the optimal solution of the required cost function.These controllers are widely used for tracking various trajectories for airship missions [29][30][31][32].
The current study makes a significant contribution to the field of airship control research.We have developed and linearized a comprehensive nonlinear mathematical model based on rigid body dynamics, validated the model through rigorous analysis, and proposed a hybrid control approach to the linearized model.The hybrid controller incorporates the dynamics of the required trajectory by employing a switching mechanism between two methods at the transition points of the trajectory profile.This novel approach effectively improves the performance of airship control by considering trajectory dynamics, resulting in satisfactory results which enhances the understanding and application of airship control strategies.
Figure 1 shows a flowchart of the work throughout the article.In "Methods" section, the airship non-linear model is derived and linearized with small disturbance theory to obtain the LTV model of the airship.The LTV model is utilized in designing LQT controllers.Two LQT control methods are presented in "Methods" section according to assumed costates transformation.The results of the hybrid LQT controller are presented in "Results" section.The controllers specifications are stated in "Discussion" section.Also, the discussion in "Discussion" section ends up with using a hybrid LQT controller in the various used trajectories.The article is concluded in "Conclusions" section.

Axes rotation transformation matrix
The rotation transformation matrix is a relation between the body axes (xyz) and fixed frame of reference (XYZ) [33] (see Fig. ) 2, which is given by, (1)

Airship mathematical modeling
The airship modeling is similar to the model of the rigid body with 6DOF.The model consists of two main parts: kinematics and kinetics.

Kinetics
Kinetics is a study of body dynamics due to force(s) or torque(s) which are applied on it considering its mass [34].The dynamics of airship are listed as follows, with considering the center of gravity (C.G.) eccentricity, where its C.G. differs from its center of volume (C.V.) at least in z-direction due to vertical thruster: a. Translational and rotational equations of motion According to Euler-Newton approach, the translational and rotational equations of motion of a rigid body can be represented by [35], where, F is the vector of external forces acting on the body, m is the body mass, a is the body acceleration, M is the vector of external moments acting on the body, and Ḣ is the rate of change of the angular momentum.The kinematic analysis of linear acceleration in Eq. 6 is valid when the axes are at C.G., but it is not the situation.So, another compensation terms are added to the translational equations of motion due to eccentric C.G., as follows Also, Eq. 4 is used in rotational equations of motion with other compensation terms due to C.G. eccentricity, as follows b.Gravity forces and moments Airship gravitational forces acting through the C.G. due to its weight were given by Bernard [36], If there is C.G. eccentricity, induced moments were developed by Hibbeler [34], as follows c.Aerodynamic forces and moments Semi-empirical aerodynamic equations were developed by Jones and DeLaurier [26] for uniform flow and extended to consider the side flow effect by Atyya et al. [27] with introducing the flow angles, namely, the angle of attack (α) , Eq. 13, and the side- slip angle (β) , Eq. 14, as shown in Fig. 3.
The aerodynamic forces and moments are given by, 3 Airship aerodynamic axes where x n is the nose position in x-direction with respect to xyz body axes shown in Fig. 3, and is given by, where, a 1 and a 2 are the hull front and rare major axes, respectively.C X , C Y , C Z , C L , C M , and C N are the aerodynamic coefficients for forces and moments.They are defined as follows, and the aerodynamic coefficients The airship propulsion system consists of three thrusters, namely, vertical thruster T z , right thruster T r , , and left thruster T l [37], as shown in Fig. 4, and given by, (15a)

Full dynamic equations
The dynamic equations of the airship according to Euler-Newton approach can be written as follows, The schematic diagram of airship non-linear model is shown in Fig. 5.The model consists of five subsystems (aerodynamic, propulsion, gravity, inertia, and kinematics), seven inputs (T r , T l , T z , δ eL , δ eR , δ rT and δ rB ) , and twelve outputs ( 16) (u, v, w, p, q, r, X, Y , Z, φ, θ and ψ) .The definitions of these symbols are listed in the "Nomenclature" section.First, the aerodynamic, the propulsion, and the gravity subsystems are solved together to calculate the inertia subsystem using Eq. 17.Then, the kinematics subsystem is fed to obtain the airship output states.

Linearization of airship non-linear model
Linearization is the most powerful technique to simplify the airship dynamic modeling to apply the linear control theory.Linearization is performed using small disturbance theory.It is assumed that any variable can be expressed by initial value and small disturbance.And the higher order terms are neglected.The following approximations will be applied on airship non-linear model variables, where, P l is the linearized variable, P 0 is an initial value, and P and Q are the small distur- bances.The following vectors will be used in the linearization process: The linearization is obtained at general operating states u 0 , v 0 , w 0 , p 0 , q 0 , r 0 , φ 0 , θ 0 , ψ 0 , α 0 , and β 0 with general nominal action of the control signals T r 0 , T l 0 , T z 0 , δ rT 0 , δ rB 0 , δ eR 0 , and δ eL 0 .These operating values are changing with time.Therefore, the linear model is considered as a linear time varying (LTV) model.The current approach differs from the traditional method where linearization is performed around a specific operating point.In this study, we have developed a general linearization model that can be applied to any operating point.Thus, the linear model is not limited to a single set of (18) P l = P 0 + P, tan θ ≈ θ, sin θ ≈ θ , cos θ ≈ 1, P 2 ≈ 0, PQ ≈ 0 Fig. 5 Open loop schematic diagram of airship non-linear model conditions but is applicable to a wide range of operating points.Consequently, the proposed airship linear model is nonautonomous.

Linearization of kinematics
The kinematics Eqs. 3 and 5 can be linearized as follows, where,

Linearization of transnational and rotational equations of motion
The transnational and rotational equations of motion with eccentric C.G. Eqs. 9 and 10 can be linearized as follows, where, (19

Linearization of gravity forces and moments
The gravity forces and moments Eqs.11 and 12 can be linearized as follows, (20a) where,

Linearization of aerodynamic forces and moments
The linearization of angle of attack (α) , Eq. 13, and side slip angle (β) , Eq. 14, are, where, The aerodynamic forces and moments, Eq. 15, can be linearized as follows, where,

Linearization of propulsive forces and moments
The propulsive forces and moments Eq. 16 can be linearized as follows, where,

Full linearized equations
We set all disturbances equal to zero in the full dynamic linear equations, to get the reference flight conditions, where, B I , B G , B A1 , B A2 and B P are defined in Eqs.20g, 21b, 24d, 24f and 25b respec- tively.So, the ariship dynamic Eq. 17, can be linearized to, where, (24v) , and A P are defined in Eqs.19a, 19b, 20a, 20b, 21a, 24a, 24c, and 25a respectively.
The nominal action can be given by adding Eqs.24f and 25b, then, we substitute Eq. 27 into Eq.29 to get, where,

Airship mass properties
The airship 3D model built by Atyya et al. [27], shown in Fig. 6, with hollow hull and tri-motors as a propulsive unit will be used in the current study.The propulsive unit position is chosen to shift airship C.G. in z direction only x G = 0, y G = 0 and z G � = 0 , (28a) Fig. 6 3D CAD model of the airship whereas the mass properties of the airship is presented in Table 1.The values provided in Table 1 are estimations based on the mass properties of the Solid-Works CAD model used in our study.It is important to note that the airship body is proposed to be constructed from wood, and the model assumes the presence of three engines along with a hypothetical static load.Figure 7 shows airship geometrical parameters on elevation and side view.

Airship optimal control tracking
Airship is modeled as a rigid body, considering the difference between C.G. and C.V. The model is linearized with small disturbance theory to get an LTV system instead of the non-linear one.Hence, the linear control theory can be applied to design the controller  which stabilizes and improves the performance of the airship while performing its tracking missions.Linear quadratic tracking (LQT) is one of the optimal control techniques with fully output feedback.The derivation of this controller is based on the theorems of optimization of calculus of variation to optimize a quadratic cost function which expresses the required target from the system [38][39][40][41].The derivation of the control signal for general LTV system is explained in the following steps: 1. Consider an LTV system, with a cost function, where, x(t) is the state vector of size n, u(t) is the control vector of size r, y(t) is the output vector of size m, A(t) is n × n state matrix, B(t) is n × r control matrix, C(t) is m × output matrix, F t f is the terminal cost weighted matrix, Q(t) is the error weighted matrix, R(t) is the control weighted matrix, and z(t) is the reference vector of size m.Note that the two matrices Q(t) and F t f should be symmetric positive semidefinite, whereas the matrix R(t) should be symmetric positive definite.

Construct the Hamiltonian equation,
where (t) is the costate vector.

Compute the optimal control signal u(t)
4. Obtain the state and costate equations, where, (32) 5.Then, a transformation of the costates is assumed.There are two assumed transformation. a.
We assume the first transformation by intuition from Eq. 37 as follows, where, P(t) and G(t) are unknown matrices of size n × n to be determined.Then, substitute the state Eq.36 and costate Eq. 37 into the transformation Eq. 40 to get, assuming the reference input has the same dynamics of the system as follows, then, and then, solve the two differential Riccati equations (DRE), backward in time with final condition and, backward in time with final condition then, the optimal control signal can be obtained as, If C = I , this leads to V (t) = W (t) .So, the two Riccati matrices P(t) and G(t) are equivalent.b.The second transformation is given by Naidu, D Subbaram [41], (39) where, P(t) is an unknown matrix of size n × n and g(t) is an unknown vector of size n to be determined.Then, substitute the state Eq.36 and costate Eq. 37 into the transformation Eq. 49, then, solve the DRE, backward in time with final condition and the non-homogeneous vector differential equation, backward in time with final condition then, the optimal control signal can be obtained as, Figure 8 shows a diagram of the closed loop system of the airship with controller.The non-linear model consists of five subsystems with seven inputs: four of (49 them are the deflections of vertical and horizontal stabilizers (δ rT , δ rB , δ eR and δ eL ) , and the others are the thrusters (T r , T l and T z ) .The non-linear model is line- arized with small disturbance theory to design the controller which minimizes the error between the output vector [u v w p q r X Y Z φ θ ψ] and the required vector

Linearized model verification
The airship non-linear model of Eq. 17 is linearized by small disturbance theory to get the linear model Eq.28.The linear model is verified with input control signal U shown in Fig. 9 with limitation [−25 • , 25 • ] on fin deflections, whereas Figs. 10 and 11 present the response of airship states for linear and non-linear models.The absolute error between the linear and non-linear models is introduced in Figs. 12 and 13.The previous figures

Comparison of tracking controllers
The two LQT controllers presented in Eqs.48 and 55 are used to improve the performance of the airship.For simplicity, new subscripts are introduced as M1 and M2 to refer to the result of the first method in Eq.48 and the second method in Eq. 55 of LQT controllers, respectively.A proposed trajectory is defined in three phases: climbing for 20 s, hovering for 20 s, and going through apart of helical shape for 110 s.Figures 14 and  15 show the responses of the two controllers, whereas Figs.16 and 17 show the control action signal of the two methods respectively.M12 is referring to it.These figures gives satisfactory results of airship trajectory following with acceptable control actions.

Airship performance for different trajectories
The airship performance is tested by three different trajectories, circular, helical, and bowed using LQT M12 tracking control method with weights in Eq. 56.These trajecto- ries define the airship position (X r , Y r and Z r ) and orientation (φ r , θ r and ψ r ) , and then the kinematic Eqs. 2 and 5 are used to determine the required airship velocity components (u r , v r and w r ) and angular rate (p r , q r and r r ) .The required yawing angle ψ r is computed by the following equation [43,44], The required position for the different trajectories is defined as follows:

Circular trajectory
The selection of the time factor in these trajectories is influenced by various factors, including the complexity of the airship model, the simulation sample time, the computational capabilities of the machine used for simulation, and the interval of convergence of the linear model.It is important to note that the choice of these specific trajectories was obtained through a process of trial and error, taking into consideration the aforementioned factors.The aim was to find a suitable time factor that allows for a controlled and manageable simulation while still capturing the essential dynamics of the system.For a real-world model, the determination of the time factor would primarily depend on the sample time of the sensors used for data acquisition.It is crucial to align the simulation parameters with the real-world conditions to ensure accurate representation and analysis.The proposed position for the circular trajectory is defined as follows,

Helical trajectory
The proposed position for the helical trajectory is defined as follows, (57) ψ r = tan −1   LQT M2 control method is applied around t = 20 s and t = 40 s , whereas LQT M1 control method is applied elsewhere.There- fore, LQT M12 control method improves the performance of the airship with a stable and an acceptable control signal.

Linearized model verification
The absolute error between the linear and non-linear models (59)  12 and 13 indicates that the linear model simulates the non-linear one with a good accuracy, since the order of error between the linear and non-linear models is 10 −3 or less, except for three states w, q, and r (see Table 2.These three states   have small biases at different parts of the trajectory.However, these biases are acceptable when compared with the absolute values of the states.Table 2 presents the error analysis between the responses of Figs. 10 and 11, which correspond to the linear and nonlinear models, respectively.The inputs for these models are taken from Fig. 9.The error in the table is calculated as the absolute difference between the actual response of the nonlinear model and the estimated response of the linear model.

Comparison of tracking controllers
Figure 16 shows that the first method resulted in a sudden change at the points of switching the trajectory phase.On the other hand, the second method had a smooth change at these points with a strong response at the beginning (see Fig. 17).Same conclusions were presented by Suiçmez [45].These analyses led to use the first method of LQT controller with switching to the second method at the points of changing trajectory profile (see Figs. 18, 19, and 20).This hybrid controller was utilized to improve the airship performance in different trajectories-circular, helical, and bowed-and gave a great results as shown in Figs.21, 22, 23, 24, 25, 26, 27, 28, and 29.

Conclusions
Our study focused on the development of a comprehensive nonlinear mathematical model for airship dynamics in a six-degree-of-freedom (6DOF), considering rigid body dynamics.By applying small disturbance theory, we derived a linearized model that served as a valuable tool for validation and control design purposes.
In terms of control design, we introduced a hybrid controller that combined two methods of linear quadratic tracking (LQT) to optimize the airship's performance.The first method, LQT M1 , demonstrated effective results in minimizing the error between the output and the desired states, while maintaining stability and generating acceptable control signals.However, it proved to be sensitive to changes in the trajectory profile.
To address this sensitivity, we incorporated the dynamics of the required trajectory into the control design using the second method, LQT M2 .By considering the transition points of the trajectory profile, we implemented a switching mechanism between LQT M1 and LQT M2 .This hybrid controller configuration yielded satisfactory results, combining the advantages of both methods.
In conclusion, our work contributes to the field of airship control by providing a comprehensive nonlinear mathematical model, validating the linearized model, and proposing a hybrid control approach that improves performance by accounting for trajectory dynamics.The findings of this study provide valuable insights for further refining and optimizing control strategies in airship applications.

F
The vector of external forces acting on the airship, N H The vector of the airship angular momentum, kg.m 2 /s M The vector of external moments acting on the airship, N.m a The airship body acceleration, m/s The half distance between right and left thrusters, m l z The distance between the airship C.V. and propulsive unit center in z-direction, m

Fig. 1
Fig. 1 Airship optimal controller design flow chart

Fig. 2
Fig. 2 Axes rotation transformation via Euler angles and C N 4 are derived by Atyya et al. [27].d.Propulsive forces and moments

Fig. 7
Fig. 7 Elevation and side view of airship CAD modeling showing different geometrical parameters

Fig. 25 Fig. 26
Fig. 25 Response of airship velocity, angular rate, and orientation to helical trajectory

Table 1
Airship mass properties z ) Maximum thickness to chord ratio of of fin airfoil (t/c) max 0.06 Minimum drag value of of fin airfoil C dmin 0.01 Derivative of fin lift coefficient with respect to the angle of ∂C l

Table 2
Statistical analysis of the absolute error between linear and non-linear models Cosine of the angle (•) C L , C M , C NCoefficients of aerodynamic moments L s , M q , N r , respectively,m 3 C X , C Y , C Z Coefficients of aerodynamic forces F s , F q , F r , respectively, m 2 C L1 C M1 , C M2 , C M3 , C M4 C N 1 , C N 2 , C N 3 , C N 4 C X1 , C X2 C Y 1 , C Y 2 , C Y 3 , C Y 4 C Z1 , C Z2 , C Z3 , C Z4 F x , F y , F zExternal forces acting on the airship in x, y and z directions, respectively, N F x,a , F y,a , F z,a Aerodynamic forces in x, y and z directions, respectively, N F x,g , F y,g , F z,gGravitational forces in x, y and z directions, respectively, N F x,p , F y,p , F z,pPropulsive forces in x, y and z directions, respectively, NI (••)Second moment of inertia about the axis (•) , or product moment of inertia about the axes (••) , kg.m2 J(t)Cost function of the LTV model L, M, N External moments acting on the airship about x, y and z axes, respectively, N.m L a , M a , N a Aerodynamic moments about x, y and z axes, respectively, N.m L g , M g , N gGravitational moments about x, y and z axes, respectively, N.m L p , M p , N p Propulsive moments about x, y and z axes, respectively, N.m Q Hull front and rare major axes, respectively, m gThe gravitational acceleration, 9.80665 m/s 2 l y