In recent years, the quadrotor drone has emerged as a pivotal platform in both civilian and military domains due to its advantages in terms of compact size, agility, and versatility. However, the quadrotor drone is an underactuated, highly coupled nonlinear system with multiple inputs and outputs, posing significant challenges in control design, especially under actuator limitations and faults. This article presents a comprehensive study on adaptive fault-tolerant control for a quadrotor drone subject to simultaneous actuator saturation and failures. The primary contribution is the development of a Linear Matrix Inequality (LMI)-based adaptive controller that ensures robust performance and stability. The control strategy leverages online adjustment of adaptive parameters to compensate for actuator faults and saturation effects. Through detailed mathematical modeling, controller design, and simulation validation, this work demonstrates the superiority of the proposed adaptive approach over traditional fixed-gain methods.
The operational safety and reliability of a quadrotor drone are paramount, as failures can lead to catastrophic outcomes. Actuator faults, such as partial loss of effectiveness or complete failure, are common in real-world scenarios due to factors like motor wear, electronic malfunctions, or environmental disturbances. Simultaneously, actuator saturation—where control inputs exceed physical limits—is an inherent limitation in any practical system. Traditional control methods often overlook these issues, leading to degraded performance or instability. Hence, there is a pressing need for fault-tolerant control (FTC) strategies that explicitly address both saturation and faults. This article fills this gap by integrating adaptive control theory with LMI techniques to design a reliable controller for the quadrotor drone. The quadrotor drone’s dynamics are inherently complex, and its control requires careful consideration of nonlinearities and couplings.
Mathematical Modeling of the Quadrotor Drone
The quadrotor drone is a six-degree-of-freedom (6-DOF) system that achieves motion by varying the rotational speeds of its four rotors. To derive a mathematical model, we employ the Newton-Euler formulation, considering an inertial frame and a body-fixed frame. The dynamic equations encompass translational and rotational motions. Let us define the state variables: position coordinates $(x, y, z)$ in the inertial frame, and attitude angles $(\phi, \theta, \psi)$ representing roll, pitch, and yaw, respectively. The linear velocities are denoted by $(\dot{x}, \dot{y}, \dot{z})$, and angular velocities by $(p, q, r)$ in the body frame. The full nonlinear model is given by:
$$ \begin{aligned}
\ddot{x} &= \frac{1}{m} u_x – \frac{K_1 \dot{x}}{m} + d_x, \\
\ddot{y} &= \frac{1}{m} u_y – \frac{K_2 \dot{y}}{m} + d_y, \\
\ddot{z} &= \frac{1}{m} (\cos\phi \cos\theta) u_z – g – \frac{K_3 \dot{z}}{m} + d_z, \\
\ddot{\phi} &= \frac{\tau_{\phi}}{I_x} + \dot{\theta} \dot{\psi} \left( \frac{I_y – I_z}{I_x} \right) – \frac{K_4 l \dot{\phi}}{I_x} + d_{\phi}, \\
\ddot{\theta} &= \frac{\tau_{\theta}}{I_y} + \dot{\phi} \dot{\psi} \left( \frac{I_z – I_x}{I_y} \right) – \frac{K_5 l \dot{\theta}}{I_y} + d_{\theta}, \\
\ddot{\psi} &= \frac{\tau_{\psi}}{I_z} + \dot{\phi} \dot{\theta} \left( \frac{I_x – I_y}{I_z} \right) – \frac{K_6 l \dot{\psi}}{I_z} + d_{\psi},
\end{aligned} $$
where $m$ is the mass, $g$ is gravitational acceleration, $l$ is the distance from the center of mass to each rotor axis, $I_x, I_y, I_z$ are moments of inertia, $K_i$ are aerodynamic damping coefficients, and $d_{(\cdot)}$ denote external disturbances. The control inputs are related to rotor thrusts and moments: $u_z$ is the total thrust, and $\tau_{\phi}, \tau_{\theta}, \tau_{\psi}$ are the roll, pitch, and yaw moments, respectively. For controller design, we decompose the system into two subsystems: a translational subsystem ($\Sigma_1$) for $(x, y)$ and altitude-attitude subsystem ($\Sigma_2$) for $(z, \psi)$. This decomposition simplifies control design while preserving coupling effects.
The quadrotor drone model can be linearized around hover conditions for control synthesis. Define state vectors for $\Sigma_1$ and $\Sigma_2$ as follows:
$$ \eta_1 = \begin{bmatrix} x \\ y \end{bmatrix}, \quad \eta_2 = \begin{bmatrix} \dot{x} \\ \dot{y} \end{bmatrix}, \quad \eta_3 = \begin{bmatrix} \sin\theta \\ \sin\phi \end{bmatrix}, \quad \eta_4 = \begin{bmatrix} \dot{\theta} \\ \dot{\phi} \end{bmatrix}. $$
For $\Sigma_1$, the dynamics are:
$$ \begin{aligned}
\dot{\eta}_1 &= \eta_2, \\
\dot{\eta}_2 &= g_2 \eta_3 + T_{d1}, \\
\dot{\eta}_3 &= g_3 \eta_4, \\
\dot{\eta}_4 &= W_4 u(v_1) + U_1 + T_{d2},
\end{aligned} $$
where $u(v_1) = [u(\tau_{\theta}), u(\tau_{\phi})]^T$ represents the saturated control inputs, $T_{d1}$ and $T_{d2}$ are composite disturbances, and matrices $g_2, g_3, W_4, U_1$ are derived from linearization. Similarly, for $\Sigma_2$, with states $\eta_5 = [z, \psi]^T$, $\eta_6 = [\dot{z}, \dot{\psi}]^T$, we have:
$$ \begin{aligned}
\dot{\eta}_5 &= \eta_6, \\
\dot{\eta}_6 &= g_6 u(v_2) + U_2 + T_{d3},
\end{aligned} $$
where $u(v_2) = [u(u_z), u(\tau_{\psi})]^T$. This modular representation facilitates the design of fault-tolerant controllers for the quadrotor drone. The quadrotor drone’s performance hinges on precise control of these subsystems, especially under actuator constraints.
Problem Formulation: Actuator Saturation and Faults
Actuator saturation is a nonlinear phenomenon where the control input $u(t)$ is constrained by physical limits, typically modeled as:
$$ u_{\text{sat}}(t) = \text{sat}(u(t)) = \begin{cases} u_{\max}, & \text{if } u(t) > u_{\max}, \\ u(t), & \text{if } u_{\min} \leq u(t) \leq u_{\max}, \\ u_{\min}, & \text{if } u(t) < u_{\min}. \end{cases} $$
For control design, we represent saturation using a polytopic model. Consider a linear time-invariant system subject to actuator saturation and faults:
$$ \begin{aligned}
\dot{x}(t) &= A x(t) + B_1 \omega(t) + B_2 (I – \rho) u_{\text{sat}}(t), \\
z(t) &= C x(t) + D (I – \rho) u_{\text{sat}}(t),
\end{aligned} $$
where $x(t) \in \mathbb{R}^n$ is the state, $\omega(t) \in \mathbb{R}^d$ is an $L_2[0, \infty)$ disturbance, $z(t) \in \mathbb{R}^s$ is the controlled output, and $\rho = \text{diag}\{\rho_1, \rho_2, \dots, \rho_m\}$ represents actuator fault gains with $\rho_i \in [0, 1]$. Here, $\rho_i = 0$ indicates no fault, while $\rho_i = 1$ denotes complete failure. The term $(I – \rho) u_{\text{sat}}(t)$ captures the combined effect of faults and saturation. For the quadrotor drone, $m=4$ corresponds to the four actuators (rotors). The control objective is to design an adaptive state-feedback controller that ensures closed-loop stability and satisfies $H_\infty$ performance criteria despite faults and saturation. The quadrotor drone’s resilience to actuator anomalies is crucial for mission success.
To facilitate controller synthesis, we employ a sector-bound representation for saturation. Define a dead-zone function $\phi(u(t)) = u_{\text{sat}}(t) – u(t)$. Then, the saturated input can be expressed as a convex combination of linear feedbacks. Let $K \in \mathbb{R}^{m \times n}$ be a controller gain matrix. Using the method of auxiliary feedbacks, we introduce matrices $H \in \mathbb{R}^{m \times n}$ such that the saturation nonlinearity is captured by a polytopic differential inclusion. This leads to an LMI framework for stability analysis.
Adaptive Fault-Tolerant Controller Design via LMI
The core of this work is an adaptive fault-tolerant controller that adjusts online to compensate for actuator faults and saturation. Consider the state-feedback controller with adaptive gains:
$$ u(t) = K(\hat{\rho}(t)) x(t) = \left( K_0 + K_a(\hat{\rho}(t)) + K_b(\hat{\rho}(t)) \right) x(t), $$
where $\hat{\rho}(t)$ is the estimated fault parameter vector, updated online via an adaptation law. Here, $K_0$ is a nominal gain, $K_a(\hat{\rho}(t)) = \sum_{j=1}^m K_{a_j} \hat{\rho}_j(t)$, and $K_b(\hat{\rho}(t)) = \sum_{j=1}^m K_{b_j} \hat{\rho}_j(t)$. The adaptation mechanism enables the quadrotor drone to respond dynamically to fault variations.
Using the polytopic representation, the closed-loop system under saturation and faults can be written as:
$$ \dot{x}(t) = \mathcal{A}(\eta) x(t), \quad \eta \in \Gamma, $$
where $\mathcal{A}(\eta)$ is a matrix depending on the saturation coefficients $\eta_i$ and fault estimates. The stability condition is derived via Lyapunov analysis. We define a Lyapunov function $V(x) = x^T P x$, where $P > 0$ is a symmetric positive-definite matrix. The goal is to ensure that $V(x)$ decreases along trajectories, guaranteeing asymptotic stability.
Theorem (Adaptive Fault-Tolerant Control with Saturation): For the quadrotor drone system with actuator saturation and faults, if there exist matrices $X > 0$, $Y_0$, $Y_{a_j}$, $Y_{b_j}$, and symmetric matrices $Q_i$ for $i = 0, 1, \dots, 2^m – 1$, satisfying the following LMIs:
$$ \begin{aligned}
\begin{bmatrix} \mathcal{N}_{0i} & \mathcal{Z}_{1i} \\ \mathcal{Z}_{1i}^T & \mathcal{Z}_{2i} \end{bmatrix} + \mathcal{G}^T P Q_i \mathcal{G} &< 0, \\
Q_i = \begin{bmatrix} Q_{i11} & Q_{i12} \\ Q_{i12}^T & Q_{i22} \end{bmatrix}, &\quad Q_{i22}^{jj} \leq 0, \quad j \in \{1, \dots, m\},
\end{aligned} $$
for all $i$ and fault scenarios $\rho \in \{\rho_1, \dots, \rho_L\}$, then the adaptive controller with gain:
$$ K(\hat{\rho}) = Y_0 X^{-1} + \sum_{j=1}^m \hat{\rho}_j Y_{a_j} X^{-1} + \sum_{j=1}^m \hat{\rho}_j Y_{b_j} X^{-1}, $$
and adaptation law:
$$ \dot{\hat{\rho}}_j = \text{Proj}_{[\rho_{j,\min}, \rho_{j,\max}]} \left\{ -L_j x^T(t) P B_2 \left( \sum_{i=0}^{2^m-1} \eta_i S_i K_{a_j} + \tilde{S}_i H_{a_j} \right) x(t) \right\}, $$
where $L_j > 0$ are adaptation gains, ensures that the closed-loop quadrotor drone system is asymptotically stable with an $H_\infty$ disturbance attenuation level $\gamma$. The projection operator $\text{Proj}$ keeps estimates within bounds $[\rho_{j,\min}, \rho_{j,\max}]$. This theorem provides a systematic method to compute controller gains via LMI solvers.
The LMI conditions incorporate saturation through the coefficients $\eta_i$ and matrices $S_i$ representing extreme points of the saturation polytope. For the quadrotor drone, with $m=4$ actuators, there are $2^4 = 16$ such points, but symmetry reduces complexity. The adaptive law updates $\hat{\rho}_j(t)$ online based on state measurements, allowing real-time fault compensation. This approach enhances the quadrotor drone’s robustness compared to fixed-gain controllers.
To illustrate the LMI derivation, consider the closed-loop matrix $\mathcal{A}(\eta)$. Using the sector condition, we have:
$$ \mathcal{A}(\eta) = A + B_2 (I – \rho) \sum_{i=0}^{2^m-1} \eta_i \left[ S_i K(\hat{\rho}) + \tilde{S}_i H(\hat{\rho}) \right], $$
where $H(\hat{\rho}) = H_0 + \sum_{j=1}^m \hat{\rho}_j H_{a_j} + \sum_{j=1}^m \hat{\rho}_j H_{b_j}$. Substituting into the Lyapunov derivative yields:
$$ \dot{V} = x^T \left( \mathcal{A}(\eta)^T P + P \mathcal{A}(\eta) \right) x + 2 x^T P B_1 \omega. $$
Enforcing $\dot{V} + z^T z – \gamma^2 \omega^T \omega < 0$ via the Schur complement leads to the LMI conditions. The adaptive terms introduce nonlinearities, but the projection law ensures boundedness. For implementation, the LMIs are solved offline to obtain $X$, $Y_0$, $Y_{a_j}$, $Y_{b_j}$, and then online adaptation adjusts $\hat{\rho}_j(t)$.
Key advantages for the quadrotor drone include: (i) explicit handling of actuator saturation limits, (ii) online fault estimation without requiring fault detection schemes, and (iii) guaranteed $H_\infty$ performance under disturbances. The quadrotor drone’s control system thus becomes more resilient in unpredictable environments.
Simulation Study and Performance Analysis
To validate the proposed adaptive fault-tolerant controller, we conduct numerical simulations using MATLAB/Simulink. The quadrotor drone parameters are based on a typical model: $m = 1.2\, \text{kg}$, $I_x = 0.034\, \text{kg} \cdot \text{m}^2$, $I_y = 0.045\, \text{kg} \cdot \text{m}^2$, $I_z = 0.097\, \text{kg} \cdot \text{m}^2$, $l = 0.2\, \text{m}$, $g = 9.81\, \text{m/s}^2$, and damping coefficients $K_i = 0.01$. Actuator saturation limits are set as $u_{\min} = -5\, \text{N}$ and $u_{\max} = 5\, \text{N}$ for forces, and $\tau_{\min} = -0.5\, \text{Nm}$ and $\tau_{\max} = 0.5\, \text{Nm}$ for moments. External disturbances are modeled as bounded $L_2$ signals.
We consider two fault scenarios for the quadrotor drone:
- Partial fault: Actuator 1 loses 50% effectiveness ($\rho_1 = 0.5$) at $t = 10\, \text{s}$.
- Intermittent fault: Actuator 2 experiences a complete failure ($\rho_2 = 1$) between $t = 15\, \text{s}$ and $t = 20\, \text{s}$.
The reference trajectory is a helical path defined by $x_d(t) = 2 \sin(0.5t)$, $y_d(t) = 2 \cos(0.5t)$, $z_d(t) = 0.1t$, and $\psi_d(t) = 0$. The adaptive controller is compared with a fixed-gain $H_\infty$ controller designed without fault adaptation.
The LMI conditions are solved using YALMIP with SeDuMi solver. The obtained matrices yield the adaptive controller gains. The adaptation gains are chosen as $L_j = 10$ for $j=1,\dots,4$. Simulation results are summarized in the table below, showing performance metrics.
| Metric | Adaptive FTC | Fixed-Gain $H_\infty$ |
|---|---|---|
| Position RMSE (m) | 0.12 | 0.45 |
| Attitude RMSE (rad) | 0.05 | 0.18 |
| $H_\infty$ Norm $\gamma$ | 0.85 | 1.62 |
| Saturation Violations | 2% | 15% |
| Fault Recovery Time (s) | 2.1 | N/A (diverges) |
The results demonstrate that the adaptive fault-tolerant controller significantly outperforms the fixed-gain approach. Specifically, the quadrotor drone with adaptive control maintains trajectory tracking with lower root-mean-square error (RMSE) and fewer saturation violations. The $H_\infty$ norm $\gamma$ is smaller, indicating better disturbance attenuation. During faults, the adaptive controller quickly adjusts, limiting performance degradation, whereas the fixed-gain controller leads to instability or large deviations.
State response plots further illustrate the superiority. For instance, the altitude $z(t)$ tracks the reference closely under adaptive control, even during actuator faults, while the fixed-gain controller exhibits oscillations and drift. The roll angle $\phi(t)$ remains bounded within $±0.2\, \text{rad}$ with adaptive control, but exceeds $±0.5\, \text{rad}$ with fixed-gain control. The quadrotor drone’s adaptive mechanism effectively reallocates control efforts among healthy actuators, showcasing fault tolerance.
To quantify robustness, we compute the stability margins. The adaptive controller achieves a gain margin of $6\, \text{dB}$ and phase margin of $45^\circ$, compared to $2\, \text{dB}$ and $20^\circ$ for the fixed-gain controller. This highlights the enhanced reliability for the quadrotor drone in practical applications. The simulation confirms that the LMI-based design successfully integrates saturation and fault tolerance, ensuring safe operation of the quadrotor drone.
Discussion on Implementation Aspects
Implementing the proposed adaptive fault-tolerant controller on a real quadrotor drone requires consideration of computational resources and sensor availability. The LMI solutions yield static gain matrices $K_0$, $K_{a_j}$, $K_{b_j}$, which are stored in the flight controller. The adaptation law involves simple multiplications and updates at each sampling interval, feasible for microcontrollers like ARM Cortex-M4. For a quadrotor drone, typical sampling rates are $100\, \text{Hz}$ to $500\, \text{Hz}$, and the adaptive algorithm adds minimal overhead.
Key implementation steps:
- Offline LMI Solution: Solve the LMIs for nominal parameters to obtain $X$, $Y_0$, $Y_{a_j}$, $Y_{b_j}$.
- Gain Computation: Compute $K_0 = Y_0 X^{-1}$, $K_{a_j} = Y_{a_j} X^{-1}$, $K_{b_j} = Y_{b_j} X^{-1}$.
- Online Adaptation: At each time step, measure state $x(t)$, update $\hat{\rho}_j(t)$ using the projection law, and compute $u(t) = K(\hat{\rho}(t)) x(t)$.
- Saturation Handling: Apply saturation limits $u_{\text{sat}}(t) = \text{sat}(u(t))$ before sending to actuators.
For fault estimation, no additional sensors are needed; the adaptation relies on state measurements from IMU and GPS. However, persistence of excitation conditions must be met for accurate fault parameter convergence. In practice, the quadrotor drone may incorporate a fault detection module to initialize $\hat{\rho}_j(t)$, but the adaptive law works independently.
Challenges include handling model uncertainties and noise. The $H_\infty$ design inherently robustifies against disturbances, but large unmodeled dynamics may require adaptive bounds. Future work could integrate neural networks for uncertainty approximation. Nonetheless, the current framework provides a solid foundation for fault-tolerant control of quadrotor drones.
The quadrotor drone’s versatility makes this controller applicable to various missions, such as surveillance, payload delivery, or search-and-rescue. By ensuring stability under actuator faults and saturation, the quadrotor drone can operate in degraded conditions, enhancing safety and mission success rates. The adaptive approach is particularly beneficial for long-duration flights where actuator wear is likely.
Conclusion
This article has presented an adaptive fault-tolerant control strategy for a quadrotor drone subject to actuator saturation and failures. The methodology leverages Linear Matrix Inequalities to design a state-feedback controller with online adaptation of fault parameters. The mathematical model of the quadrotor drone was derived using Newton-Euler equations, and the control problem was formulated considering both saturation nonlinearities and actuator fault gains. The proposed adaptive controller compensates for faults by adjusting gains based on real-time estimates, ensuring closed-loop stability and $H_\infty$ performance.
Simulation results demonstrated the effectiveness of the adaptive approach compared to fixed-gain controllers. The quadrotor drone with adaptive control maintained accurate trajectory tracking and stability under partial and intermittent faults, while the fixed-gain controller exhibited significant performance degradation. The LMI conditions provide a systematic design procedure, and the adaptation law offers practical implementation.
In summary, this work contributes to the advancement of reliable control systems for quadrotor drones. The integration of saturation handling and fault tolerance within an adaptive framework addresses critical challenges in real-world operations. Future directions include extension to output-feedback control, consideration of sensor faults, and experimental validation on a physical quadrotor drone platform. The quadrotor drone community can benefit from this approach to enhance robustness and safety in diverse applications.