When a physical actuator degrades or fails entirely—whether it is a jammed flap on an aircraft, a damaged quadrotor propeller, or a failing motor on a robotic arm—traditional control systems often fail. Standard controllers are typically designed with passive robustness, meaning they use fixed parameters calculated to tolerate a narrow margin of unexpected disturbances. However, when a severe hardware failure occurs, these passive margins are quickly overwhelmed, leading to instability and catastrophic failure.
Active Fault-Tolerant Control (FTC) shifts the paradigm from passive endurance to active adaptation. Instead of ignoring the fault, an active FTC system dynamically reconfigures its control strategy in real time. By treating hardware degradation as a dynamic variable rather than a catastrophic anomaly, these systems preserve stability and tracking performance under severe operational stress.
The Core Intuition: Faults as Dynamic Uncertainties
To understand why active FTC works, we must look at how it mathematically frames a physical failure. In a classical framework, an actuator failure is seen as a structural change to the system's equations. Active FTC bypasses this complexity by re-framing actuator loss-of-effectiveness or bias as a time-varying uncertainty or an additive external disturbance.
As highlighted by Shuang Zhang et al. in A Single Parameter-Based Adaptive Approach to Robotic Manipulators With Finite Time Convergence and Actuator Fault, active adaptation yields significantly faster recovery speeds and lower tracking errors than passive alternatives. By continuously adjusting control parameters on-the-fly, the system counteracts performance drops near-instantaneously, even when actuators lose the majority of their torque capacity.
This concept of treating faults as generalized disturbances allows controllers to maintain performance without complex diagnostic logic. For instance, Hongde Qin et al. demonstrate in Disturbance-Observer-Based Prescribed Performance Fault-Tolerant Trajectory Tracking Control for Ocean Bottom Flying Node that a high-performance disturbance observer can absorb the power loss of a thruster seamlessly. Because the controller views the fault simply as an external force to be countered, it avoids the delay and complexity of isolating exactly which component failed.
[ Physical Actuator Fault ]
│
▼
[ Modeled as Time-Varying Uncertainty / Disturbance ]
│
▼
[ Real-Time Adaptive Loop / Disturbance Observer ] ──► [ Instantaneous Control Correction ]
Fault-Agnostic vs. Explicit Estimation
Within active FTC, researchers utilize two primary strategies to handle the math of a failing component: explicit fault estimation and fault-agnostic adaptation.
Explicit Fault Estimation
This approach actively calculates the exact magnitude and timing of the failure. For high-stakes systems like aerial vehicles, speed is paramount. Jeffrey Mao et al. show in From Propeller Damage Estimation and Adaptation to Fault Tolerant Control: Enhancing Quadrotor Resilience that combining L1 adaptive control with real-time optimization allows a quadrotor to estimate severe rotor damage in under 0.2 seconds. This rapid estimation allows the control law to aggressively cancel out asymmetric forces before the vehicle enters an unrecoverable spin.
Fault-Agnostic Adaptation
Conversely, identifying the exact failure state can sometimes introduce unnecessary computational overhead or dangerous control signal "chattering." In Adaptive Fault-Tolerant Guaranteed Performance Control for Euler-Lagrange Systems With Its Application to a 2-Link Robotic Manipulator, Deqiang Cheng demonstrates that a system can stabilize itself without real-time identification of specific fault parameters. By utilizing adaptive laws paired with Barrier Lyapunov Functions, the controller directly counteracts the energy of unknown faults. This fault-agnostic architecture suppresses chattering and stabilizes mechanical movements within very short timeframes, maintaining tracking precision an order of magnitude higher than traditional robust methods.
Dynamic Control Allocation and Reconfiguration
When an actuator does not merely degrade but fails completely, the system must alter how it interacts with the physical world. This is achieved through dynamic control allocation in over-actuated systems—systems that possess more actuators than strictly required to control their degrees of freedom.
When a failure occurs, the control allocation layer acts as a dynamic router. It stops sending commands to the broken component and recalculates how to distribute the desired control forces among the remaining healthy actuators.
Desired Trajectory ──► [ High-Level Controller ] ──► Desired Forces/Torques
│
▼
[ Control Allocation Mixer ]
/ │ \
▼ ▼ ▼
Actuator 1 Actuator 2 [Actuator 3] (FAILED)
▲ ▲ │
│ │ X (Isolated)
(Increased (Increased
Workload) Workload)
This self-configuring capability is highly apparent in complex robotic platforms. For example, Marko Nonhoff et al. demonstrate in Economic model predictive control for snake robot locomotion that an economic model predictive control framework allows a snake robot to automatically "discover" new, non-sinusoidal movement patterns when a joint fails. Rather than relying on a pre-programmed recovery strategy for every single joint, the optimization framework naturally redistributes the workload to healthy actuators to maintain speed and efficiency.
A similar mechanism applies to aerospace applications. In Helicopter Safe Landing Trajectory after Main Rotor Actuator Failures, Yunjie Wang et al. show that when a main rotor actuator becomes stuck, safe landings can be achieved purely by reconfiguring the control mixer logic. By adjusting how commands map to the remaining operational control surfaces, the system compensates for the physical jam without requiring any structural or hardware modifications to the aircraft.
Key Foundational Work
Joint Synthesis of Controller and Fault Detector and Its Application to Motor Drive Control System (2001) — contributed 1 concept elements to this direction.
Towards a Fault Tolerant AHS Design (1995) — contributed 1 concept elements to this direction.