Sliding Mode Control

The sliding surface is a user-defined manifold in the state space where the system exhibits desired dynamics. Once the state trajectory reaches this surface, it slides along it toward the origin. The surface is typically designed as a linear combination of the tracking error and its derivatives: s = ė + λe. The parameter λ determines the convergence rate during sliding mode. Proper surface design ensures asymptotic stability and prescribed transient performance independent of plant parameters.

The control signal switches at infinite frequency across the sliding surface to counteract deviations. The law takes the form u = -K * sign(s), where sign(s) is the signum function. This discontinuous switching creates a chattering phenomenon in practice. The gain K must be chosen larger than the maximum bound of the matched uncertainty to guarantee the reaching condition s * ṡ < -η|s|, ensuring finite-time convergence to the surface.

SMC provides complete insensitivity to parameter variations and external disturbances that enter the system through the same channel as the control input. These are called matched uncertainties. During sliding mode, the system's behavior is governed solely by the sliding surface dynamics, not the plant parameters. This robustness property is the primary advantage over linear controllers like PID, making SMC ideal for systems with significant modeling errors or unpredictable load disturbances.

Ideal SMC requires infinite switching frequency, which is physically impossible. In practice, the discontinuity causes high-frequency oscillations called chattering, which can excite unmodeled dynamics and damage actuators. Mitigation strategies include:

• Boundary layer approach: Replacing sign(s) with a saturation function sat(s/φ) to smooth the control signal
• Higher-order SMC: Driving s and its derivatives to zero, hiding the discontinuity under an integrator
• Super-twisting algorithm: A second-order SMC that produces continuous control while retaining robustness

Before sliding occurs, the system is in the reaching phase, where it moves from an arbitrary initial condition toward the sliding surface. The reaching law dictates this transient behavior. Common designs include:

• Constant rate reaching: ṡ = -K * sign(s)
• Exponential reaching: ṡ = -K * sign(s) - q * s, which accelerates convergence when far from the surface
• Power rate reaching: ṡ = -K * |s|^α * sign(s) for 0 < α < 1, providing fast convergence near the surface A well-designed reaching law minimizes the reaching time while avoiding excessive control effort.

Stability in SMC is proven using Lyapunov's direct method. A candidate Lyapunov function V = ½s² is chosen. The control law is designed to satisfy the η-reachability condition: V̇ = s * ṡ ≤ -η|s|, guaranteeing finite-time convergence to the sliding surface. Once on the surface, the reduced-order sliding dynamics are designed to be asymptotically stable. This rigorous mathematical foundation distinguishes SMC from heuristic control methods.

An optimization-based control strategy that uses an explicit dynamic process model to predict future plant behavior over a finite receding horizon. Unlike SMC's reactive switching, MPC proactively solves a constrained optimization problem at each time step to compute the optimal control sequence, making it ideal for multi-variable systems with hard input and state constraints.

A model-agnostic strategy that shares SMC's goal of insensitivity to disturbances but achieves it through estimation rather than switching. An Extended State Observer (ESO) treats both internal non-linearities and external forces as a 'total disturbance' and cancels it in real-time, eliminating the need for a precise mathematical model or high-frequency control chattering.

A critical practical consideration in SMC implementation. The theoretical infinite-frequency switching is impossible in real actuators, causing high-frequency oscillations. Key mitigation techniques include:

• Boundary Layer Saturation: Replacing the discontinuous signum function with a continuous saturation or hyperbolic tangent approximation.
• Higher-Order Sliding Modes: Driving the sliding variable and its derivatives to zero to hide the discontinuous term under an integrator.

The mathematical backbone for proving SMC convergence. A Lyapunov function V(x) is a positive-definite energy-like function of the state error. The control law is designed to guarantee that the derivative of V is negative definite, ensuring the system state is driven to the sliding surface in finite time and remains there, guaranteeing asymptotic stability despite matched uncertainties.

A popular second-order sliding mode controller that completely eliminates the need for the derivative of the sliding variable. It consists of a continuous control term and a discontinuous integral term, producing a continuous output that masks the switching. This makes it highly suitable for mechanical actuators where direct chattering would cause wear, while retaining finite-time convergence and robustness.

A complementary strategy where controller gains are automatically adjusted based on a scheduling variable (e.g., operating point, error magnitude). In the context of SMC, adaptive laws can dynamically scale the discontinuous gain to be just large enough to overcome the current disturbance bound, minimizing unnecessary chattering while preserving robustness across the full non-linear operating range.