Control System Fundamentals

Control Systems Engineering — Industrial Reference

01 — Foundation

Control System Fundamentals

A control system manages, commands, and regulates the behaviour of other systems. Every industrial process relies on feedback loops to maintain desired output despite disturbances.

▶ Diagram 1 — Open Loop vs Closed Loop Control

???? Closed-Loop Gain

T(s) = G(s) / [1 + G(s)H(s)] G(s) — forward path gain H(s) — feedback path gain

The fundamental relationship governing every feedback control system. The denominator (1 + GH) is the characteristic equation.

⚡ Signal Types

Step: R(s) = 1/s Ramp: R(s) = 1/s² Parabola:R(s) = 1/s³ Impulse: R(s) = 1

Standard test signals used to characterise system response: steady-state error, transient, settling.

???? System Types

Type 0: finite Kp, zero Kv Type 1: finite Kv, zero Ka Type 2: finite Ka

Number of pure integrators (1/s poles) in the open-loop transfer function determines the system type and steady-state error.

????

Laplace Transform converts differential equations into algebraic equations: ℒ{f(t)} = F(s) = ∫₀^∞ f(t)e⁻ˢᵗ dt — enabling transfer function analysis in the s-domain.

Transient Response Specifications

⏱️

Rise Time

10% → 90% of final value. t_r ≈ 1.8/ωn

????

Peak Overshoot

Mp = e^(-πζ/√(1-ζ²)) × 100%

????

Settling Time

t_s ≈ 4/(ζωn) for ±2% criterion

????

Steady-State Error

e_ss = lim(s→0) s·E(s)

02 — Core Controller

PID Controller — Deep Dive

The Proportional-Integral-Derivative controller is the most widely deployed controller in process industries — over 95% of industrial control loops use some form of PID.

▶ Diagram 2 — PID Controller Internal Structure

⚙️ Proportional (P)

u = Kp × e(t) Effect: reduces error Risk: steady-state offset Tuning: increase Kp → faster response, more overshoot

Rise Time ReductionHigh

Steady-State ErrorMedium

∫ Integral (I)

u = Ki ∫e(t)dt Effect: eliminates SS error Risk: integral windup, reduced stability Tuning: increase Ki → slower, eliminates offset

Error EliminationComplete

Stability ImpactReduces

∂ Derivative (D)

u = Kd × de/dt Effect: predicts future error Risk: noise amplification Tuning: increase Kd → damps oscillation, improves stability

Damping EffectHigh

Noise SensitivityHigh

Ziegler-Nichols Tuning Rules

Controller	Kp	Ti (=Kp/Ki)	Td (=Kd/Kp)	Use Case
P	0.5 × Ku	∞	0	Simple, offset accepted
PI	0.45 × Ku	Pu / 1.2	0	No noise, eliminate offset
PID	0.6 × Ku	Pu / 2	Pu / 8	Fastest response with stability
PD	0.8 × Ku	∞	Pu / 8	Noisy environments, no integrator

Ku = Ultimate gain (oscillation onset), Pu = Ultimate period of sustained oscillation

MATLAB / Simulink Implementation

%% PID Controller Design — MATLAB Example
s = tf(‘s’);

% Define Plant: second-order system with delay
Gp = tf(10, [1 3 10]);

% PID controller gains
Kp = 1.8;  Ki = 0.9;  Kd = 0.5;

% Transfer function form
Gc = Kp + Ki/s + Kd*s;

% Closed-loop transfer function
T = feedback(Gc*Gp, 1);

% Plot step response
step(T); grid on;
title(‘PID Closed-Loop Step Response’);

% Get performance metrics
info = stepinfo(T);
fprintf(‘Rise Time: %.3f s | Overshoot: %.2f%%\n’, info.RiseTime, info.Overshoot);

⚠️

Integral Windup — When the actuator saturates, the integrator continues accumulating error, causing large overshoot when the setpoint is reached. Always implement anti-windup clamping in real systems.

✅

Industrial Best Practice — Use derivative-on-measurement (not error) to avoid derivative kick during setpoint changes. Apply a first-order low-pass filter: Gd(s) = Kd·s/(τf·s+1)

03 — Frequency Analysis

Frequency Domain Methods

Bode plots, Nyquist diagrams, Nichols charts, and Root Locus are the primary tools for analysing stability margins, bandwidth, and frequency-domain performance of control systems.

???? Bode Plot — Gain & Phase Margins

Magnitude (dB): |G(jω)|_dB = 20·log₁₀|G(jω)| First-order lag 1/(τs+1): corner freq ωc = 1/τ → -20 dB/dec Second-order 1/(s²/ωn²+2ζs/ωn+1): slope: -40 dB/dec resonant peak at ωr = ωn√(1−2ζ²) Pure delay e^{-Ls}: magnitude = 0 dB (always) phase = −Lω rad (phase lag grows!)

GM
Gain Margin
Amount gain can increase before instability. Measured at ωpc (phase = −180°). GM > 6 dB recommended.
PM
Phase Margin
Phase above −180° at ωgc (gain = 0 dB). PM > 45° → well-damped, PM ≈ 60° → optimal.
BW
Bandwidth ωBW
Frequency where |T(jω)| drops to −3 dB. Higher BW = faster closed-loop response.
SM
Sensitivity Peak Ms
Max of |S(jω)| = 1/|1+L|. Ms < 2 (6 dB) for robust stability.

▶ Bode Plot — Phase & Gain Margin Annotated

???? Gain Margin (GM)

At phase crossover freq ωpc (∠G = -180°): GM = −20·log₁₀|G(jωpc)| dB Interpretation: GM = 6 dB → gain can double before instability GM = 20 dB → gain can increase ×10 before instability GM < 0 dB → already unstable!

⚠️

Rule of thumb: GM > 6 dB minimum; > 10 dB for robust industrial loops.

???? Phase Margin (PM)

At gain crossover freq ωgc (|G| = 0 dB): PM = 180° + ∠G(jωgc) Damping ratio approximation: ζ ≈ PM / 100 (for PM in degrees) PM = 45° → ζ ≈ 0.45 (slightly underdamped) PM = 60° → ζ ≈ 0.60 (well-damped, optimal) PM = 90° → ζ ≈ 0.90 (overdamped, slow)

✅

Design target: PM between 45°–70° for good transient response.

???? Nyquist Plot & Stability Criterion

▶ Nyquist Plot — Stable System (No Encirclement of −1)

Nyquist Stability Criterion: Z = P − N Z = closed-loop RHP poles P = open-loop RHP poles N = clockwise encirclements of −1 For stability: Z = 0 → N = P (stable plant: P=0 → N must = 0) Gain/Phase margin from Nyquist: GM: distance along real axis from −1 to curve crossing PM: angle to unit circle crossing

M-CIRCLES — CONSTANT |T(jω)|

M-circles are loci of constant closed-loop gain magnitude drawn on the Nyquist plane. They are circles centred on the real axis at −M²/(M²−1) with radius M/(M²−1).

M value	dB	Centre (Re)	Meaning
0.707	−3 dB	−1.0	−3 dB bandwidth
1.0	0 dB	−0.5	Equal gain locus
1.3	+2.3 dB	−2.0	Typical peaking limit
2.0	+6 dB	−1.33	Excessive resonance

???? Nichols Chart — with M-circles & N-circles

▶ Nichols Chart — Open-loop dB vs Phase with Constant M/N Contours

Nichols Chart axes: X-axis: Open-loop phase ∠G(jω) (degrees) Y-axis: Open-loop magnitude |G(jω)| (dB) M-circles (constant closed-loop gain): Centre = −M²/(M²−1) on real axis Radius = M/(M²−1) Tangent point → resonant peak M_p N-circles (constant closed-loop phase): Circle centre at (−½, N/2) N = tan(phase angle of T(jω)) Stability margins on Nichols chart: GM: horizontal dist at 0 dB to −180° line PM: vertical dist at −180° to 0 dB line

NICHOLS vs BODE vs NYQUIST

Method	Axes	GM/PM	Best For
Bode	\|G\| dB, ∠G vs ω	Direct read	Design, loop shaping
Nyquist	Im vs Re of G(jω)	Geometric	RHP poles, MIMO
Nichols	\|G\|dB vs ∠G	Direct read	M-circles, closed-loop

????

Nichols chart advantage: The M-circle tangent to the open-loop locus directly gives the closed-loop resonant peak M_p — no separate calculation needed. The tangent M-circle radius = M_p.

???? Root Locus — Rules & Asymptote Construction

▶ Root Locus + Asymptote Angles (3-pole, 1-zero system)

ROOT LOCUS CONSTRUCTION RULES

1
Branches = Poles
Number of branches = number of open-loop poles n. Each starts at a pole (K=0) and ends at a zero or infinity (K→∞).
2
Real Axis Rule
Locus on real axis to the left of an odd total count of real poles + real zeros.
3
Asymptote Angles
For n−m branches going to infinity:

φ_k = (2k+1)·180° ───────────────── n − m k = 0, 1, 2, …, (n−m−1) Example (n=3, m=1 → 2 asymptotes): φ₀ = 1×180°/2 = 90° φ₁ = 3×180°/2 = 270°
4
Asymptote Centroid σa
σa = Σ(real poles) − Σ(real zeros) ───────────────────────── n − m
5
Breakaway / Break-in
Points where locus leaves/enters real axis. Solve dK/ds = 0 on real-axis segment.
6
Imaginary Axis Crossing
Use Routh-Hurwitz on characteristic polynomial to find gain K where poles hit jω-axis → stability boundary.

????

Nyquist Stability Criterion (full statement) — Z = P − N, where Z = number of closed-loop RHP poles, P = number of open-loop RHP poles, N = net clockwise encirclements of −1+j0. A stable closed-loop system requires Z = 0. For a stable open-loop plant (P = 0), the Nyquist plot must have zero net encirclements of the −1 point.

Stability: Z = P − N = 0 Sensitivity: S(s) = 1/(1+L(s)) → disturbance rejection Complementary: T(s) = L(s)/(1+L(s)) → tracking/noise Waterbed: S(s) + T(s) = 1 ← fundamental constraint

04 — Modern Control

State-Space Representation

State-space is the foundation of modern control theory — enabling multi-input multi-output (MIMO) systems, controllability, observability, and optimal control design.

???? State-Space Equations

ẋ(t) = Ax(t) + Bu(t) ← State eq. y(t) = Cx(t) + Du(t) ← Output eq. Where: x(t) — state vector (n×1) u(t) — input vector (m×1) y(t) — output vector (p×1) A — system matrix (n×n) B — input matrix (n×m) C — output matrix (p×n) D — feedthrough (p×m)

???? Transfer Function ↔ State-Space

TF from SS: G(s) = C(sI−A)⁻¹B + D Characteristic equation: det(sI−A) = 0 (eigenvalues of A = poles) Example: G(s) = 1/(s²+3s+2) A = [[0,1],[-2,-3]] B = [[0],[1]], C=[1,0], D=0

✅ Controllability

???? = [B | AB | A²B | … | Aⁿ⁻¹B] rank(????) = n → Fully controllable (can drive state to any point)

Test using rank(ctrb(A,B)) in MATLAB. Essential for pole placement and LQR design.

???? Observability

???? = [C; CA; CA²; …; CAⁿ⁻¹] rank(????) = n → Fully observable (state can be inferred from y)

Required for state estimator (Luenberger observer or Kalman filter) design. Check with rank(obsv(A,C)).

???? Pole Placement

u = -Kx (state feedback) ẋ = (A – BK)x + Br Place eigenvalues of (A-BK) at desired pole locations MATLAB: K = place(A, B, poles)

Choose poles for desired transient response. System must be controllable.

LQR — Linear Quadratic Regulator (Optimal Control)

Minimise cost function: J = ∫₀^∞ [xᵀQx + uᵀRu] dt Solution via Riccati equation: AᵀP + PA − PBR⁻¹BᵀP + Q = 0 Optimal gain: K* = R⁻¹BᵀP MATLAB: [K, P, e] = lqr(A, B, Q, R)

????

Q matrix — weights state deviations (larger Q → aggressive correction of states)

⚡

R matrix — weights control effort (larger R → more conservative actuator use, less energy)

Kalman Filter — Optimal State Estimator

▶ Kalman Filter Predict-Update Cycle

05 — Stability Analysis

Stability Analysis Methods

A system is BIBO stable if every bounded input produces a bounded output. Equivalently, all closed-loop poles must lie in the left-half s-plane (LHP).

????

Routh-Hurwitz

Algebraic test — count sign changes in first column of Routh array

????

Nyquist

Encirclement of -1+j0 point determines stability via encirclement rule

????

Bode Margins

GM > 6 dB and PM > 45° ensures robust stability with adequate margins

????

Lyapunov

V(x) > 0 and V̇(x) < 0 proves asymptotic stability for nonlinear systems

Routh-Hurwitz — Worked Example

Characteristic equation: s⁴ + 2s³ + 3s² + 4s + 5 = 0 Routh Array: s⁴ | 1 3 5 s³ | 2 4 0 s² | 1 5 0 (= 3-4/2, 5-0/2) s¹ | -6 0 0 (= 4-10/1) s⁰ | 5 Sign changes in col 1: 2 changes → 2 RHP poles → UNSTABLE

????

Routh Special Cases: (1) Zero in first column — replace with ε>0 and evaluate. (2) Entire row zeros — use auxiliary polynomial (pole on jω-axis).

Necessary condition for stability: All coefficients must be: • positive (or all negative) • present (no missing terms) Sufficient: Routh test (all first column elements same sign)

Stability vs Response — s-Plane Pole Location Map

▶ s-Plane — Pole Locations & System Behaviour (no overlapping)

06 — Digital Implementation

Digital Control Systems

Modern control is implemented digitally on PLCs, DSPs, and embedded systems. Continuous-time designs must be discretised and sampling effects carefully managed.

z-Transform Pairs

f(t) / f[k]	F(s)	F(z)
Impulse δ[k]	1	1
Step u[k]	1/s	z/(z−1)
Ramp kT·u[k]	1/s²	Tz/(z−1)²
e^(-akT)	1/(s+a)	z/(z−e^{-aT})
sin(ωkT)	ω/(s²+ω²)	z·sin(ωT)/(z²-2zcos(ωT)+1)

Discretisation Methods

ZOH
Zero-Order Hold
Most common. Exact mapping for piecewise constant inputs. Use c2d(G,'zoh')
Tus
Tustin (Bilinear)
s ≈ (2/T)(z−1)/(z+1). Preserves frequency response shape up to Nyquist.
FWD
Forward Euler
s ≈ (z−1)/T. Simple but less accurate. Can cause instability at large T.

Shannon Sampling Theorem & Nyquist Rate

Nyquist criterion: fs ≥ 2 × fmax (sampling freq ≥ 2× signal bandwidth) Practical rule for control: fs = 10× to 20× closed-loop bandwidth → Ts = 1/fs (sampling period) Anti-aliasing: Low-pass filter before ADC cutoff = fs/2 = Nyquist frequency

%% Digital PID — C-code implementation (embedded/PLC)
typedef struct {
    float Kp, Ki, Kd;
    float prev_error, integral;
    float Ts;          // sampling period
    float out_min, out_max;  // anti-windup limits
} PID_t;

float PID_update(PID_t* pid, float setpoint, float measured) {
    float error      = setpoint – measured;
    float derivative = (measured – pid->prev_error) / pid->Ts; // on measurement
    pid->integral  += error * pid->Ts;

    // Anti-windup clamping
    if (pid->integral > pid->out_max) pid->integral = pid->out_max;
    if (pid->integral < pid->out_min) pid->integral = pid->out_min;

    float output = pid->Kp*error + pid->Ki*pid->integral – pid->Kd*derivative;
    pid->prev_error = measured;
    return clamp(output, pid->out_min, pid->out_max);
}

07 — Industrial Applications

Industrial Control Architecture

Real process industries use layered automation architectures — from field sensors up through PLCs, SCADA systems, and enterprise MES/ERP integration.

▶ Diagram 3 — Industrial Automation Hierarchy (ISA-95 / Purdue Model)

Industrial Protocols & Standards

???? Field Protocols

HART 4-20mA PROFIBUS PA/DP FOUNDATION Fieldbus Modbus RTU/TCP DeviceNet CANopen

HART allows digital comms overlaid on 4-20mA loops — most widely used for smart transmitter configuration.

???? Industrial Ethernet

PROFINET IRT EtherNet/IP EtherCAT Modbus TCP POWERLINK OPC-UA

OPC-UA is the modern standard for secure, platform-independent data exchange — foundation of IIoT integration.

???? Safety Standards

IEC 61508 IEC 61511 SIL 1-4 LOPA SIS / SIL

Safety Integrity Level (SIL) quantifies the required risk reduction of a safety function in process industries.

Common Industrial Control Loops

Loop Type	Controlled Variable	Typical Sensor	Actuator	Controller
Flow	Flow Rate	Coriolis, DP, Vortex	Control Valve	PI (fast, noisy)
Level	Tank Level	DP, Radar, Ultrasonic	Pump, Valve	PID (P or PI common)
Pressure	Process Pressure	DP/gauge transmitter	Pressure regulator	PID (tight tuning)
Temperature	Temp / Heat Transfer	PT100, Thermocouple	Heater, Steam valve	PID (slow process)
Cascade	Primary + Secondary	Multiple sensors	Multiple actuators	Master-Slave PID
Ratio	A/B flow ratio	Two flow meters	Two control valves	Ratio controller

08 — Advanced & Modern

Advanced Control Strategies

Beyond PID — modern control strategies that handle constraints, nonlinearities, uncertainty, and multi-variable interactions critical in today’s complex industrial processes.

???? Model Predictive Control (MPC) — Industry Standard for MIMO

Optimisation at each step k: min Σ [‖y(k+i|k)−r‖²_Q + ‖Δu(k+i)‖²_R] u i=1..N subject to: x(k+1) = Ax(k) + Bu(k) (model) u_min ≤ u ≤ u_max (input limits) y_min ≤ y ≤ y_max (output limits) |Δu| ≤ Δu_max (rate limits) Apply only u*(k), re-solve at k+1 (Receding Horizon Principle)

✓
Handles Constraints Explicitly
Input/output/rate constraints built into optimisation — crucial for actuator limits.
✓
MIMO Coordination
Handles interactions between multiple inputs and outputs simultaneously.
✓
Feedforward of Known Disturbances
Future setpoint changes and measurable disturbances included in prediction horizon.

Honeywell RMPCT Aspen DMCplus Shell ADOC ABB MPC

???? Fuzzy Logic Control

Rules example (temp control): IF error is NB AND Δerror is NB THEN output is PB IF error is Z AND Δerror is Z THEN output is Z

Encodes expert knowledge as linguistic rules. Excels where mathematical model is unavailable. Used in washing machines to power plants.

• Centroid (centre of gravity) — most common
• Mean of Maximum (MOM)
• Bisector — median of area
• Smallest/Largest of Maximum

???? Adaptive Control

MRAC — Model Reference: e = y_m − y_p θ̇ = −γ·e·φ (MIT rule) Self-Tuning Regulator: Estimate → Design → Apply

Controller parameters adjust online to handle time-varying processes, gain scheduling for aircraft flight envelope, pH control.

• MRAC (Model Reference Adaptive Control)
• STR (Self-Tuning Regulator)
• Gain Scheduling
• L1 Adaptive Control

????️ Robust Control (H∞)

H∞ norm: ‖T_zw‖∞ = sup_ω σ_max[T(jω)] H∞ design: min_K ‖F_l(P, K)‖∞ < γ Guarantees stability for all plants in uncertainty set Δ

Synthesises controllers that guarantee stability and performance despite structured/unstructured uncertainty. Used in aerospace, robotics.

μ (structured singular value) extends H∞ to handle structured uncertainty. D-K iteration alternates between H∞ synthesis and μ analysis to minimise peak μ value.

???? AI-Driven Control — Latest Industrial Trends

???? Reinforcement Learning Control

Agent learns policy π(s)→a: Q(s,a) ← Q + α[r + γ max_a’ Q(s’,a’) − Q] Applications: • HVAC energy optimisation • Chemical reactor control • Data centre cooling (DeepMind) • Robotic manipulation

Deep RL (PPO, SAC, TD3) now deployed in real industrial processes. Sim-to-real transfer using digital twins is key industrial enabler.

???? Physics-Informed Neural Control

Neural ODE / PINN: dx/dt = f_θ(x, u, t) Loss = L_data + λ·L_physics Hybrid model: ẋ = f_physics(x) + f_NN(x, u)

Neural networks trained with physical constraints — mass balance, energy conservation. Dramatically reduces data requirements for process control.

Controller Selection Guide

Controller	MIMO	Constraints	Nonlinear	Tuning Effort	Best Application
PID	❌	❌	❌	Low	SISO loops, flow, pressure
Cascade PID	Partial	❌	❌	Medium	Fast inner / slow outer loops
MPC	✅	✅	Partial	High	Refinery, polymer, pharma
Fuzzy Logic	Partial	Soft	✅	Medium	No model, expert knowledge
LQR/LQG	✅	❌	❌	Medium	Aerospace, robotics (linear)
H∞ Robust	✅	❌	❌	Very High	Uncertainty, aerospace, precision
RL / AI	✅	✅	✅	Very High	Complex, data-rich, dynamic envs

⚡ Quick Reference — Critical Formulas

Second-Order System: G(s) = ωn² / (s² + 2ζωn·s + ωn²) Natural freq: ωn = √(K/M) Damping ratio:ζ = C/(2√KM) Damped freq: ωd = ωn√(1−ζ²) ζ<1: underdamped (oscillatory) ζ=1: critically damped (fastest, no OS) ζ>1: overdamped (slow, no OS)

Final Value Theorem (steady-state): y_ss = lim(t→∞)y(t) = lim(s→0) s·Y(s) Initial Value: y(0⁺) = lim(s→∞) s·Y(s) Sensitivity function: S(s) = 1/(1+L(s)) → disturbance rejection T(s) = L(s)/(1+L(s)) → tracking, noise S + T = 1 ← waterbed effect!

Control System Fundamentals