Q5Control System

(a) Optimal Control

Optimal control theory addresses the problem of determining the control input u(t) for a dynamic system, described typically by state equations ẋ = f(x,u,t), that minimizes (or maximizes) a specified performance index (cost functional) J, subject to the system dynamics as equality constraints and possibly additional inequality constraints on states/inputs. A widely used performance index in linear systems is the quadratic cost:

where Q and R are positive semi-definite/positive-definite weighting matrices that penalize state deviation and control effort respectively. For a linear time-invariant plant ẋ = Ax + Bu, minimizing this cost leads to the Linear Quadratic Regulator (LQR), whose optimal control law is a state feedback u = -Kx, where K = R⁻¹BᵀP and P is the unique positive-definite solution of the Algebraic Riccati Equation:

Optimal control formalizes the trade-off between fast/accurate regulation (large Q) and conservative, low-effort actuation (large R), and is derived using variational calculus (Pontryagin's Minimum Principle) or dynamic programming (Bellman's Principle of Optimality/Hamilton-Jacobi-Bellman equation). It is applied extensively in aerospace guidance, robotics trajectory optimization, and economic/process control where minimizing energy, time, or fuel consumption is as important as achieving the desired state.

(b) Non-linear Control

Non-linear control deals with systems whose governing differential equations do not satisfy the superposition and homogeneity properties assumed by classical linear control theory (transfer functions, Laplace transforms, Bode/root-locus techniques do not directly apply). Common sources of non-linearity include saturation, dead-zone, backlash, hysteresis, and Coulomb friction in actuators, as well as intrinsically non-linear plant dynamics (e.g. pendulum dynamics involving sin θ, or bilinear/product terms between states). Because closed-form transfer functions generally do not exist, non-linear systems are analyzed and controlled using dedicated techniques such as: phase-plane analysis and the describing function method (for approximate frequency-domain analysis of mild non-linearities and limit-cycle prediction); Lyapunov's direct method, which establishes stability by finding an energy-like function V(x) that is positive definite with a negative semi-definite derivative along system trajectories, without needing to solve the non-linear differential equations explicitly; feedback linearization, which uses a non-linear state transformation and non-linear inner-loop control law to cancel the plant's non-linear terms and render the closed-loop dynamics linear, so that standard linear design can then be applied to the outer loop; and sliding-mode control, which drives system trajectories onto a designed switching surface and keeps them there despite model uncertainty or external disturbances, offering strong robustness at the cost of high-frequency chattering in the control signal. Non-linear control is essential in robotics, aircraft flight control, power electronics converters, and any system where linearized small-signal models fail to capture large-signal or highly non-linear plant behavior.

A key qualitative distinction from linear systems is that non-linear systems can exhibit behaviors with no linear counterpart, such as finite escape time (the state diverging to infinity in finite time), multiple isolated equilibrium points, and self-sustained limit cycles (bounded periodic oscillations that persist independent of initial conditions, unlike the marginally stable oscillations of a linear system which depend entirely on the initial energy injected). Because linear stability tools (root locus, Bode, Nyquist) are built on the assumption of a single, globally valid transfer function, they can only be applied to a non-linear plant after linearizing about a specific operating point (using a Jacobian/small-signal approximation), and such a linearized model is only locally valid near that operating point — this is precisely why genuinely non-linear analysis and design techniques, as summarized above, remain indispensable wherever a system must operate correctly across a wide range of conditions rather than in the small neighborhood of a single set-point.