Survey on Flight Control Technology for Large-Scale Helicopter
Table 1
Summary on modern control methodologies.
Control methods
Advantages
Disadvantages
LQR/LQG Control
Guaranteed stability margins for state feedback LQR; good transient response and disturbance rejection.
Full state feedback requires significant usage of sensors; Kalman filter degrades the frequency domain properties.
Eigenstructure Assignment Control
Close-loop performance index can be designed directly; the control method is relatively simple to implement without requiring high order dynamic compensators.
An analytical linear plant is required; stability and robustness is not guaranteed; dynamic cross-coupling between inputs and outputs cannot be canceled.
Adaptive Control
Control gains can be adjusted automatically using on-line adaption; robustness is achieved within bounded system uncertainties; the regulated output dynamics are allowed to be non-minimum phase.
Parameter drift is caused by process noise; high frequency flexible modes are difficult to control; large rates of adaption are practically non-realizable.
LPV Control
Nonlinear natures of the plant are captured accurately using LPV modeling; controller robustness is improved over conventional gain-scheduled controllers.
The complexity of LPV controller is significantly increased; system uncertainties caused by rapid varying flight parameters cannot be completely eliminated.
Sliding Mode Control
Different control laws can be adopted corresponding to the changes of system trajectories; the controller performance is insensitive to modeling errors and parameter uncertainties.
Chattering is induced by frequent controller switches; highly nonlinear sliding mode surface results in difficulties of analyzing system stability and reachability.
Backstepping Control
The exponential stability of dynamic errors guarantees asymptotic tracking of the desired command.
Calculating the derivative of pseudo control inputs results in huge computation.
MPC
External constraints are incorporated into the design process naturally; improved transient response is achieved using prediction models.
Close-loop property depends on the performance of reference model; on-line optimization is time-consuming.