In robust control theory, managing unmodeled dynamics and external disturbances often requires aggressive control actions. Sliding Mode Control (SMC) is a classic example of this approach: it uses a discontinuous switching function, such as the signum function ($\text{sign}(s)$), to instantly drive a system's state back to a predefined sliding manifold.
In an ideal mathematical world with infinite actuator bandwidth, this instantaneous switching works perfectly. However, physical hardware cannot switch at an infinite frequency. Actuator delays, sampling limits, and physical inertia cause the controller to repeatedly overshoot and undershoot the sliding manifold. This high-frequency oscillation is known as chattering. Chattering introduces severe mechanical wear, thermal stress on electrical components, and acoustic noise, creating a fundamental tension between mathematical precision and physical hardware longevity.
To bridge this gap, researchers use continuous approximations to smooth out the control signal, trading absolute asymptotic convergence for hardware safety and operational smoothness.
The Mechanism: Smooth Boundary Layers
The most direct way to mitigate chattering is to replace the hard, discontinuous signum function with a continuous or smooth approximation. This modification establishes a "boundary layer" around the sliding surface. Outside this layer, the controller acts with maximum robust gain; inside it, the control signal transitions smoothly to zero as the tracking error diminishes.
Discontinuous (Signum) Continuous Approximation (e.g., Tanh)
^ ^
| +1 | +1
-----+----- ...--+--...
| / | \
------+------ / | \
| -1 / | \
v v v -1
Several mathematical functions are commonly used to achieve this smoothing effect:
- Saturation Functions: A piecewise continuous function that acts linearly within a narrow band around the sliding surface and saturates at $\pm 1$ outside of it. Shilin Zuo in Neural Network-based Sliding Mode Control for Permanent Magnet Synchronous Motor demonstrates that replacing traditional binary switching with continuous saturation functions dampens output spikes and mechanical vibrations in electric motors without losing the ability to track desired trajectories.
- Hyperbolic Tangent Functions: The $\tanh(s/\epsilon)$ function provides an infinitely differentiable, smooth curve where the parameter $\epsilon$ controls the width of the boundary layer. Xi Ma et al. (Attitude control of rigid body with inertia uncertainty and saturation input) and Ying Xiong et al. (Robust tracking control of variable swept-wing near space vehicle based on disturbance observers) utilize hyperbolic tangent functions to eliminate high-frequency chattering in torque commands, preventing the mechanical degradation of aerospace control surfaces.
- Sigmoid Functions: Similar to the hyperbolic tangent, sigmoid functions smoothly interpolate between control extremes. En‐Chih Chang et al. (Parallel-Connected Inverter Control Method in Material Micromachining) show that substituting the signum function with a continuous sigmoid function reduces high-frequency switching noise, minimizing thermal stress on power transistors in sensitive micromachining applications.
Alternative Smoothing Pathways: Filtering and Adaptive Gains
Beyond direct function substitution, chattering can be suppressed by altering how the switching logic interacts with the physical system.
Dynamic Filtering via Integration
Instead of applying the discontinuous switching function directly to the control input, engineers can apply it to the derivative of the control signal. Because integration acts as a natural low-pass filter, integrating this discontinuous derivative yields a smooth, continuous control signal.
Yiran Shi et al. (Pitch angle control with fault diagnosis and tolerance for wind turbine generation systems) leverage this mechanism to suppress chattering in wind turbine pitch systems, protecting massive mechanical components from high-frequency wear. Similarly, Yuren Li et al. (Cascaded fast terminal sliding mode control for UAV electric braking system) embed low-pass filtering into the switching logic to prevent premature hardware failure in high-torque electric braking motors.
Adaptive and Fuzzy Gain Tuning
Fixed boundary layers can sometimes over-smooth the control law when the system is far from the target, slowing down the response time. To fix this, the reaching law parameters can be tuned dynamically based on real-time error states. Gang Liu (Coordinated Control Strategy for Stability Control and Trajectory Tracking with Wheel-Driven Autonomous Vehicles Under Harsh Situations) demonstrates that using a fuzzy controller to adaptively scale sliding-mode reaching coefficients smooths out torque adjustments, improving passenger comfort and vehicle stability on slippery roads.
The Precision-Smoothness Trade-Off
If continuous approximations protect hardware and eliminate chattering, why not use them universally with large boundary layers? The answer lies in a fundamental mathematical trade-off between tracking precision and control smoothness.
A pure discontinuous control law guarantees asymptotic or finite-time stability, meaning the tracking error converges exactly to zero. When you smooth the discontinuity, you alter the system's behavior inside the boundary layer. Within this layer, the high robust gain drops off, meaning the controller can no longer completely overcome small, persistent disturbances.
Consequently, the system shifts from absolute finite-time stability to practical stability or finite-time boundedness. The system state no longer reaches the exact target but instead settles into a small, bounded neighborhood around it.
Discontinuous Control Continuous Approximation
(Absolute Stability) (Practical Stability / Boundedness)
| |
| +---+---+ <- Boundary Layer
========X======== <- Error = 0 |...X...| <- Residual Error
| +---+---+
| |
Bing Yi et al. (Fixed-time adaptive fuzzy control for time-varying systems based on nominal substitution) explicitly formalize this tension: attempting to force the tracking error interval to absolute zero eliminates the continuity of the switching mechanism, which immediately reintroduces severe actuator chattering.
Haibin Sun et al. (Adaptive fault-tolerant controller design for airbreathing hypersonic vehicle with input saturation) also show that trading strict asymptotic stability for bounded tracking motion is a necessary compromise in practical engineering. By accepting a small, user-defined tracking error window, the control system generates fluid, continuous signals that preserve physical hardware while maintaining practical stability.
Key Foundational Work
Robust Adaptive Control for Strict-Feedback Nonlinear Systems with Non-parametric Uncertainties (2002) — contributed 2 concept elements to this direction.
An Ontology for Transitions in Physical Dynamic Systems (1998) — contributed 2 concept elements to this direction.