IITIAN1997 STUDIO

Process Dynamics & Control

Model it, close the loop, tune it.

Open the interactive studio →

Process models and Laplace transforms, interacting tanks, block diagrams, P/PI/PID control, control valves, Routh and Bode stability, Ziegler–Nichols, Cohen–Coon and IMC tuning, cascade, ratio and feedforward from the ASTU CHE241501 syllabus.

What is inside

  • Transfer functions, tanks and dynamics
  • Feedback loops, offset and PID modes
  • Routh, Bode, gain and phase margins
  • ZN, Cohen–Coon, IMC tuning and cascade

Syllabus: ASTU CHE241501 Process Dynamics and Control, semester 5, NEP 2020 (2024-25 onwards). Official syllabus PDF.

Introduction to process control and process dynamics

Topics: Motivation and importance; components of a control system; development of a mathematical model; degrees of freedom; Laplace transforms; transfer functions; forcing functions; response of first- and higher-order systems; first-order systems in series; interacting and non-interacting processes; poles and zeros.

Why control, the parts of a loop and dynamic models

Plants must run safely, meet product specifications, respect environmental limits and use energy and raw materials economically, all while feeds, ambient conditions and demands keep changing. Process control keeps chosen variables (temperatures, pressures, levels, flows, compositions) at their set points despite these disturbances. A control loop has a sensor and transmitter that measure the controlled variable, a controller that compares it with the set point and computes an output, and a final control element (usually a control valve) that changes a manipulated variable acting on the process.

Dynamic models come from unsteady-state conservation laws: accumulation = in − out + generation for mass, components and energy, plus constitutive relations (valve equations, rate laws, heat-transfer laws). The degrees of freedom (number of variables minus number of independent equations) must be zero for the model to be solvable; each control loop uses up one degree of freedom by fixing a variable at its set point. Models are usually non-linear, so they are linearised about a steady state using a Taylor series and written in deviation variables (value minus steady-state value), which makes all initial conditions zero.

The Laplace transform F(s) = ∫₀^∞ f(t)e^(−st)dt turns linear differential equations into algebraic ones: d f/dt → sF(s) − f(0). The ratio of output to input transforms with zero initial deviation is the transfer function G(s). Standard forcing functions are the step (1/s), ramp (1/s²), impulse (1) and sinusoid (ω/(s² + ω²)). After solving for Y(s), partial fractions and a table of transforms give y(t); the final-value theorem y(∞) = lim(s→0) sY(s) gives the new steady state directly.

  • Loop: sensor/transmitter → controller → final control element → process.
  • Model: accumulation = in − out + generation.
  • Deviation variables make initial conditions zero.
  • L{df/dt} = sF(s) − f(0); final value = lim sY(s).

Example: A tank of area 2 m² has outflow q = 0.5√h (m³/min, h in m) and steady inflow 0.5 m³/min. Linearise and find the transfer function H(s)/Q_in(s).

  1. Steady state: 0.5 = 0.5√h_s ⇒ h_s = 1 m.
  2. q ≈ q_s + (0.5/(2√h_s))(h − h_s) = q_s + 0.25h′ ⇒ R = 1/0.25 = 4 min/m².
  3. A dh′/dt = q′_in − h′/R ⇒ τ = AR = 8 min.
  4. H(s)/Q_in(s) = R/(τs + 1) = 4/(8s + 1).

Answer: H/Q_in = 4/(8s + 1): gain 4 m per m³/min, time constant 8 min.

Example: Use the final-value theorem to find the steady change of y for Y(s) = 2/(s(5s + 1)) (a step through a first-order system).

  1. y(∞) = lim(s→0) s × 2/(s(5s + 1)).
  2. = lim 2/(5s + 1) = 2.

Answer: y rises by 2 units.

First-order, higher-order and interacting systems; poles and zeros

A first-order system τ dy/dt + y = Kx has G(s) = K/(τs + 1): the gain K is the steady change of output per unit input and τ the time to reach 63.2 % of a step change. Liquid level, mixing tanks, thermometers and stirred heaters are first order. Two first-order systems in series give a second-order overdamped response with an S-shape (no inflection at t = 0); N in series approach a pure delay.

Two tanks are non-interacting when the first discharges freely into the second (G = R₂/((τ₁s + 1)(τ₂s + 1))) and interacting when they are connected at the bottom so the flow between them depends on both levels; the characteristic polynomial becomes τ₁τ₂s² + (τ₁ + τ₂ + A₁R₂)s + 1, whose roots are further apart, so the response is more sluggish. A second-order system with inertia, τ²ÿ + 2ζτẏ + y = Kx, can oscillate when ζ < 1 (manometers, control loops). Dead time (transport lag) e^(−θs) delays the response without changing its shape.

The roots of the denominator of G(s) are the poles; they fix the shape of the response: negative real poles give decaying exponentials, complex poles oscillations, poles with positive real parts instability. Roots of the numerator are zeros; a zero in the right half-plane gives an inverse response (the output first moves the wrong way, as in a boiler drum level or some reactors).

  • First order: K/(τs + 1).
  • Non-interacting: product of first-order terms; interacting: extra A₁R₂ term.
  • Second order: K/(τ²s² + 2ζτs + 1).
  • Poles fix stability and shape; RHP zeros cause inverse response.

Example: Two identical tanks (A = 1 m², R = 1 min/m²) are in series. Compare the transfer functions for non-interacting and interacting arrangements.

  1. Non-interacting: H₂/Q = 1/((s + 1)(s + 1)) = 1/(s² + 2s + 1): poles at −1, −1.
  2. Interacting: H₂/Q = 1/(s² + 3s + 1): poles at −0.382 and −2.618.
  3. The slower pole (−0.382, time constant 2.62 min) dominates the interacting response.

Answer: Interacting tanks respond more slowly (dominant τ ≈ 2.6 min vs 1 min).

Example: Where are the poles and zeros of G(s) = (1 − 2s)/((s + 1)(3s + 1)), and what does the step response look like?

  1. Poles: s = −1 and s = −1/3 (both negative: stable).
  2. Zero: s = +0.5 (right half-plane).
  3. A RHP zero means the response first moves in the opposite direction before settling.

Answer: Stable, with an inverse (wrong-way) initial response.

Process dynamics and control

A first-order system responds to a step by reaching 63.2 % of the final change after one time constant τ, 95 % after 3τ and 98 % after 4τ. Controllers: on–off (simple, cycles), proportional (leaves an offset), PI (integral removes offset), PID (derivative adds anticipation). Proportional band PB = 100/Kc %.

Feedback control corrects after an error appears; feed-forward measures a disturbance and acts before it affects the output. Control valves are air-to-open (fail closed) or air-to-close (fail open), chosen for safety.

  • 63.2 % at τ; 95 % at 3τ
  • P leaves offset; I removes it
  • PB = 100/Kc %

Example: A thermometer with τ = 5 s is plunged into hot water. What fraction of the change does it show after 15 s?

  1. 15 s = 3τ.
  2. 1 − e⁻³ ≈ 0.95 = 95 %.

Answer: 95 %

Example: A proportional controller has a proportional band of 50 %. Find its gain.

  1. Kc = 100/PB.
  2. = 100/50 = 2.

Answer: 2

Control keeps variables at set points despite disturbances, for safety, quality, environment and economy. A loop: sensor/transmitter, controller, final control element, process. Models from unsteady balances, linearised in deviation variables; degrees of freedom. Laplace transforms give transfer functions; standard inputs: step, ramp, impulse, sinusoid. First-order K/(τs + 1); first-order systems in series; non-interacting vs interacting tanks; second order with damping ratio; dead time. Poles decide stability and shape; RHP zeros cause inverse response.

  • Each control loop uses one degree of freedom.
  • Linearise by Taylor series about the steady state.
  • Interacting systems are slower than non-interacting ones.
  • Dead time e^(−θs) is the hardest dynamic element to control.

Key formulas

  • L{df/dt} = sF(s) − f(0)
  • G(s) = K/(τs + 1)
  • G(s) = K/(τ²s² + 2ζτs + 1)
  • Interacting: τ₁τ₂s² + (τ₁ + τ₂ + A₁R₂)s + 1
  • y(∞) = lim(s→0) sY(s)

Practice questions

The Laplace transform of a unit step is

  1. 1
  2. 1/s
  3. 1/s²
  4. s
Show the answer

Answer: B. 1/s Standard pair.

A first-order system reaches what fraction of its final change in one time constant?

  1. 50 %
  2. 63.2 %
  3. 86.5 %
  4. 95 %
Show the answer

Answer: B. 63.2 % 1 − e⁻¹.

Compared with non-interacting tanks, interacting tanks respond

  1. faster
  2. more sluggishly
  3. identically
  4. with oscillation
Show the answer

Answer: B. more sluggishly Poles spread apart.

A right-half-plane zero produces

  1. instability
  2. inverse response
  3. dead time
  4. offset
Show the answer

Answer: B. inverse response Wrong-way initial move.

Deviation variables are used so that

  1. models become non-linear
  2. initial conditions are zero
  3. gains vanish
  4. units cancel
Show the answer

Answer: B. initial conditions are zero Simplifies Laplace transforms.

Theory questions

Derive the transfer function of a liquid-level tank with a linear outlet resistance and give its step response. (5 marks)

Show the model answer

Mass balance (constant density): A dh/dt = q_in − q, with q = h/R (linear or linearised resistance). Steady state: 0 = q_in,s − h_s/R. Subtract to get deviation variables: A dh′/dt = q′_in − h′/R ⇒ AR dh′/dt + h′ = Rq′_in. Laplace with zero initial deviation: (τs + 1)H(s) = RQ_in(s), τ = AR ⇒ H/Q_in = R/(τs + 1). Step of size M: H(s) = RM/(s(τs + 1)) ⇒ h′ = RM(1 − e^(−t/τ)): first order, gain R, reaching 63.2 % of RM at t = τ.

Distinguish between interacting and non-interacting systems with an example and their transfer functions. (5 marks)

Show the model answer

Non-interacting: tank 1 discharges freely (e.g. over the top or through a long pipe to atmosphere) into tank 2, so tank 2’s level does not affect tank 1’s outflow: H₂/Q = R₂/((τ₁s + 1)(τ₂s + 1)), τ_i = A_iR_i. Interacting: the tanks are connected by a pipe at the bottom; the flow between them is (h₁ − h₂)/R₁, so each level affects the other: H₂/Q = R₂/(τ₁τ₂s² + (τ₁ + τ₂ + A₁R₂)s + 1). The added A₁R₂ term moves the roots apart; the response is slower and the effective time constants differ more, making the system harder to control.

Feedback control systems

Topics: Open- and closed-loop control systems; development of block diagrams; block-diagram reduction; servo and regulator problems; transient response of closed-loop control systems.

Block diagrams, closed-loop transfer functions, servo and regulator problems

In open-loop (feedforward or manual) control the manipulated variable is set without measuring the result; disturbances and model errors go uncorrected. In closed-loop (feedback) control the controlled variable is measured and the error drives the controller. A block diagram shows each element as a transfer function: controller G_c, final control element G_v, process G_p, disturbance path G_d and measuring element G_m, joined by summing junctions and take-off points.

Blocks in series multiply; blocks in parallel add; a feedback loop G/(1 + GH) replaces a forward path G with negative feedback H. For the standard loop Y = [G_cG_vG_p/(1 + G_OL)]Y_sp + [G_d/(1 + G_OL)]D, with the open-loop transfer function G_OL = G_cG_vG_pG_m. The servo problem studies set-point changes with D = 0; the regulator problem studies load (disturbance) changes with Y_sp = 0. The characteristic equation 1 + G_OL = 0 is the same for both and decides stability.

For a first-order process with proportional control, the closed loop is still first order but faster (τ/(1 + K)) and has a steady offset 1/(1 + K) for a unit set-point step, where K = K_cK_vK_pK_m. With PI control the closed loop becomes second order with no offset; its damping depends on K_c and τ_I. Higher-order processes under P control show overshoot and can become unstable at high gain.

  • Series: multiply; parallel: add; loop: G/(1 + GH).
  • Servo: Y/Y_sp = G_cG_vG_p/(1 + G_OL).
  • Regulator: Y/D = G_d/(1 + G_OL).
  • P control of a first-order process: offset 1/(1 + K).

Example: A process G_p = 2/(5s + 1) is under P control with K_c = 2 (G_v = G_m = 1). Find the closed-loop time constant and the offset for a unit set-point step.

  1. K = K_cK_p = 4.
  2. Y/Y_sp = (4/5)/((5/5)s + 1) = 0.8/(s + 1).
  3. τ_cl = 5/(1 + 4) = 1 min; y(∞) = 0.8 ⇒ offset = 1 − 0.8 = 0.2.

Answer: τ_cl = 1 min (five times faster), offset 0.2.

Example: Reduce the block diagram: G₁ and G₂ in series inside a unity-feedback loop, with H = 0.5 in the feedback path.

  1. Forward path G = G₁G₂.
  2. Closed loop = G/(1 + GH) = G₁G₂/(1 + 0.5G₁G₂).

Answer: Y/R = G₁G₂/(1 + 0.5G₁G₂).

Open loop acts without measurement; closed loop measures and corrects. Block diagrams: series multiply, parallel add, loop G/(1 + GH). Closed loop: Y = [G_cG_vG_p/(1 + G_OL)]Y_sp + [G_d/(1 + G_OL)]D; servo (set-point) and regulator (load) problems share the characteristic equation 1 + G_OL = 0. P control of a first-order process: faster, with offset 1/(1 + K); PI removes offset but can oscillate.

  • Negative feedback reduces sensitivity to process changes.
  • Offset falls as gain rises, but stability limits the gain.
  • The measurement lag G_m sits in the feedback path.
  • Disturbances enter through G_d.

Key formulas

  • Y/Y_sp = G_cG_vG_p/(1 + G_cG_vG_pG_m)
  • Y/D = G_d/(1 + G_OL)
  • offset (P, servo) = 1/(1 + K)
  • τ_cl = τ/(1 + K)

Practice questions

The servo problem studies response to

  1. load changes
  2. set-point changes
  3. measurement noise
  4. valve failure
Show the answer

Answer: B. set-point changes Set point.

Blocks in series combine by

  1. addition
  2. multiplication
  3. division
  4. subtraction
Show the answer

Answer: B. multiplication G₁G₂.

For P control of a first-order process with K = 4, the offset to a unit set-point step is

  1. 0
  2. 0.2
  3. 0.25
  4. 0.8
Show the answer

Answer: B. 0.2 1/(1 + 4).

The characteristic equation of a feedback loop is

  1. G_OL = 0
  2. 1 + G_OL = 0
  3. G_OL = 1
  4. 1 − G_OL = 0
Show the answer

Answer: B. 1 + G_OL = 0 Denominator of closed loop.

Proportional control of a first-order process makes the response

  1. slower
  2. faster
  3. unchanged
  4. oscillatory
Show the answer

Answer: B. faster τ/(1 + K).

Theory questions

Derive the closed-loop transfer functions for set-point and load changes of a feedback loop. (6 marks)

Show the model answer

Signals: E = Y_sp − B, B = G_mY; controller output P = G_cE; manipulated variable M = G_vP; Y = G_pM + G_dD. Substituting: Y = G_pG_vG_c(Y_sp − G_mY) + G_dD ⇒ Y(1 + G_cG_vG_pG_m) = G_cG_vG_pY_sp + G_dD. Servo (D = 0): Y/Y_sp = G_cG_vG_p/(1 + G_OL). Regulator (Y_sp = 0): Y/D = G_d/(1 + G_OL), G_OL = G_cG_vG_pG_m. Both share 1 + G_OL in the denominator.

Show that proportional control of a first-order process leaves an offset and calculate it. (5 marks)

Show the model answer

With G_p = K_p/(τs + 1), G_c = K_c, G_v = G_m = 1: Y/Y_sp = K/(τs + 1 + K) = [K/(1 + K)]/[(τ/(1 + K))s + 1], K = K_cK_p. For a unit step, y(∞) = lim s·(1/s)·K/(τs + 1 + K) = K/(1 + K), so offset = 1 − K/(1 + K) = 1/(1 + K). It falls as K_c increases but never vanishes; integral action is needed to remove it. For a unit load step with G_d = K_p/(τs + 1), y(∞) = K_p/(1 + K).

Controllers and control action; control valves

Topics: Types of controller; characteristics of proportional, PI, PD, PID and on–off controllers; control valve construction, sizing and characteristics.

Controller modes

An on–off controller switches the final element fully open or shut when the error changes sign (with a small dead band to prevent chattering). It is cheap and used for domestic heaters, refrigerators and non-critical level control, but the controlled variable cycles continuously.

A proportional controller makes the output proportional to the error: p = p̄ + K_c e, or G_c = K_c. Its strength is often given as proportional band PB = 100/K_c (%). It responds immediately but leaves an offset. Integral action adds (K_c/τ_I)∫e dt: it keeps moving the output as long as any error remains and therefore removes offset, but adds phase lag that makes the loop more oscillatory, and can wind up when the valve saturates. Derivative action adds K_cτ_D de/dt: it reacts to the rate of change of the error, anticipating it, adds phase lead and damping, but amplifies measurement noise, so it is filtered and used mainly on slow, smooth loops such as temperature.

The ideal PID controller is G_c = K_c(1 + 1/(τ_Is) + τ_Ds). Typical choices: flow loops PI (fast, noisy, no D), level loops P or PI (often loosely tuned for averaging), pressure PI, temperature and composition PID.

  • On–off: cycles; P: offset; I: removes offset; D: anticipates, adds damping.
  • PB = 100/K_c %.
  • PID: G_c = K_c(1 + 1/τ_Is + τ_Ds).
  • Flow PI, level P/PI, temperature PID.

Example: A temperature controller has a proportional band of 40 % on a 0–200 °C transmitter. What is K_c (output % per °C)?

  1. K_c (dimensionless) = 100/PB = 2.5 % output per % span.
  2. Span = 200 °C ⇒ 1 °C = 0.5 % of span.
  3. K_c = 2.5 × 0.5 = 1.25 % output per °C.

Answer: K_c = 2.5 (dimensionless), i.e. 1.25 % output per °C.

Example: A PI controller (K_c = 2, τ_I = 4 min) sees a constant error of 1 % starting at t = 0. Find its output change after 6 min.

  1. Proportional part: K_ce = 2 × 1 = 2 %.
  2. Integral part: (K_c/τ_I)∫e dt = (2/4) × 1 × 6 = 3 %.
  3. Total = 5 %.

Answer: 5 % (and it keeps rising while the error persists).

Control valves: construction, characteristics and sizing

A control valve consists of a body (globe, ball, butterfly, diaphragm) with a plug and seat that vary the flow area, and an actuator (usually a pneumatic diaphragm with spring, or piston, electric or hydraulic) driven by a 0.2–1 bar signal through a positioner. The spring decides the fail-safe action: air-to-open (fail-closed) or air-to-close (fail-open), chosen so the plant goes to a safe state on air or power failure (e.g. fuel valves fail closed, cooling-water valves fail open).

The inherent characteristic is flow versus lift at constant pressure drop: linear (f = x), equal percentage (f = R^(x−1), each equal increment of lift gives an equal percentage change of flow; rangeability R typically 20–50) and quick opening (large change near the seat, used for on–off). Installed in a line, the valve’s pressure drop falls as flow rises, distorting the characteristic: an equal-percentage valve tends to become nearly linear installed, which is why it is the common choice.

Liquid valves are sized by the flow coefficient: Q = C_v√(ΔP/SG) (US gpm, psi) or Q = K_v√(ΔP/SG) (m³/h, bar), C_v = 1.156K_v. The valve is chosen to be about 60–80 % open at design flow and is given a pressure drop of about 25–50 % of the system’s dynamic loss. Gases, flashing liquids and cavitation need special equations.

  • Fail-safe: air-to-open = fail-closed.
  • Linear, equal %, quick opening.
  • Q = K_v√(ΔP/SG); C_v = 1.156K_v.
  • Design at 60–80 % open.

Example: Water flows at 30 m³/h with a 1 bar drop across a control valve. Find K_v and C_v, and the lift for an equal-percentage valve of C_v 60 (R = 50).

  1. K_v = 30√(1/1) = 30 m³/h; C_v = 1.156 × 30 = 34.7.
  2. f = 34.7/60 = 0.578.
  3. x = 1 + ln 0.578/ln 50 = 1 − 0.548/3.912 = 0.86.

Answer: C_v ≈ 34.7; about 86 % lift (a linear valve would be 58 % open).

On–off control cycles; P gives immediate action but offset; I removes offset but adds lag and windup; D anticipates and damps but amplifies noise; PID G_c = K_c(1 + 1/τ_Is + τ_Ds); PB = 100/K_c. Control valves: body, plug, actuator, positioner; fail-safe action; inherent characteristics linear, equal percentage, quick opening; installed characteristic; sizing Q = K_v√(ΔP/SG), C_v = 1.156K_v.

  • Anti-reset windup is needed when the valve saturates.
  • Derivative on measurement avoids kicks on set-point changes.
  • Equal % valves suit loops with large pressure-drop changes.
  • Positioners improve valve accuracy and speed.

Key formulas

  • p = p̄ + K_c[e + (1/τ_I)∫e dt + τ_D de/dt]
  • PB = 100/K_c
  • f = R^(x−1) (equal %)
  • Q = K_v√(ΔP/SG)

Practice questions

Integral action is added mainly to

  1. speed up the response
  2. eliminate offset
  3. reduce noise
  4. increase damping
Show the answer

Answer: B. eliminate offset Reset.

Derivative action responds to

  1. the error
  2. the integral of error
  3. the rate of change of error
  4. the set point only
Show the answer

Answer: C. the rate of change of error Anticipation.

A proportional band of 50 % corresponds to a gain of

  1. 0.5
  2. 2
  3. 50
  4. 5
Show the answer

Answer: B. 2 100/50.

A fuel-gas valve to a furnace is usually

  1. fail-open
  2. fail-closed
  3. fail-last
  4. manual
Show the answer

Answer: B. fail-closed Safe state.

In an equal-percentage valve, equal increments of lift give

  1. equal flow increments
  2. equal percentage changes in flow
  3. no change near the seat
  4. on–off action
Show the answer

Answer: B. equal percentage changes in flow Logarithmic.

Theory questions

Compare P, PI, PD and PID controllers. (6 marks)

Show the model answer

P: output ∝ error; fast, simple; leaves offset; higher gain reduces offset but risks oscillation. PI: adds integral action; eliminates offset; slower, more oscillatory, integral windup possible; most common (flow, pressure, level). PD: adds derivative; anticipates and adds damping so higher gain can be used; still has offset; noise-sensitive. PID: combines all; no offset, good speed and damping; used for slow processes with little noise (temperature, composition); needs three parameters tuned. Response to a step load: P — offset; PI — no offset, oscillation; PID — no offset, less overshoot, faster settling.

Explain the inherent and installed characteristics of control valves and how a valve is selected. (5 marks)

Show the model answer

Inherent: flow vs lift at constant ΔP — linear (flow ∝ lift), equal percentage (dF/dx ∝ F, F = R^(x−1)), quick opening (most change at low lift). Installed: in a real line, as flow rises the line loses more pressure and the valve’s ΔP falls, so a linear valve behaves like quick opening and an equal % valve becomes nearly linear. Selection: choose the characteristic that gives a nearly linear installed gain (equal % when the valve takes a small share of total ΔP, linear when it takes most of it, quick opening for on–off), size with C_v at design flow so it is 60–80 % open, check rangeability, cavitation and noise, choose body type, materials and fail-safe action.

Stability: time- and frequency-domain analysis

Topics: Stability criteria; Routh’s criterion; root locus; frequency response analysis; gain margin, phase margin and crossover frequency; Bode plot; Nyquist plot.

Routh’s criterion and root locus

A linear closed loop is stable if every root of its characteristic equation 1 + G_OL(s) = 0 has a negative real part, so every transient decays. Routh’s test decides this without solving for the roots. Write the polynomial a_ns^n + … + a₀; all coefficients must be positive. Form the Routh array: the first two rows are the alternate coefficients; each further element is (b₁a₃ − a₁b₃)/b₁ etc. The system is stable if all elements of the first column are positive; the number of sign changes equals the number of roots in the right half-plane.

Routh’s test also gives the ultimate gain K_u: the controller gain at which an element of the first column becomes zero; the loop is then marginally stable and oscillates with the ultimate frequency ω_u, found from the auxiliary equation of the row above. For three first-order lags in series under P control, 1 + K/((τ₁s + 1)(τ₂s + 1)(τ₃s + 1)) = 0 gives K_u = a₂a₁/a₃ − 1, with a₃ = τ₁τ₂τ₃, a₂ = τ₁τ₂ + τ₂τ₃ + τ₁τ₃, a₁ = τ₁ + τ₂ + τ₃. A root locus plots how the closed-loop roots move in the s-plane as K_c goes from 0 to ∞; it starts at the open-loop poles and ends at the zeros or at infinity along asymptotes, and crosses the imaginary axis at K_u.

  • Necessary: all coefficients positive.
  • Stable ⇔ first column all positive.
  • K_u from a zero row; ω_u from the auxiliary equation.
  • Root locus: closed-loop roots vs K_c.

Example: Find the ultimate gain and period for K/((s + 1)(0.5s + 1)(0.2s + 1)).

  1. Characteristic: 0.1s³ + 0.8s² + 1.7s + 1 + K = 0.
  2. Routh s¹ row: (0.8 × 1.7 − 0.1(1 + K))/0.8 = 0 ⇒ K_u = 12.6.
  3. Auxiliary equation 0.8s² + 13.6 = 0 ⇒ ω_u = √17 = 4.12 rad/min ⇒ P_u = 2π/4.12 = 1.52 min.

Answer: K_u = 12.6, P_u = 1.52 min.

Example: Is s³ + 2s² + 3s + 10 = 0 stable?

  1. Rows: s³: 1, 3; s²: 2, 10.
  2. s¹: (2 × 3 − 1 × 10)/2 = −2; s⁰: 10.
  3. First column 1, 2, −2, 10: two sign changes.

Answer: Unstable, with two roots in the right half-plane.

Frequency response: Bode and Nyquist

If a stable linear system is forced by a sinusoid, its output settles to a sinusoid of the same frequency with an amplitude ratio AR = |G(jω)| and phase φ = ∠G(jω). For a first-order lag AR = K/√(1 + ω²τ²) and φ = −tan⁻¹ωτ; for dead time AR = 1 and φ = −ωθ; for products of elements the ARs multiply and the phases add. A Bode diagram plots log AR and φ against log ω; a Nyquist plot traces G(jω) in the complex plane as ω goes from 0 to ∞.

Bode stability criterion: if the open-loop AR exceeds 1 at the crossover frequency ω_co where the phase is −180°, the closed loop is unstable. The gain margin GM = 1/AR(ω_co) and the phase margin PM = 180° + φ at the frequency where AR = 1 measure how far the loop is from instability; typical design values are GM 1.7–2 and PM 30–45°. The Nyquist criterion is more general (it handles open-loop unstable systems and multiple crossings): the number of encirclements of −1 by the Nyquist plot determines the number of unstable closed-loop roots. Dead time is easy to handle in the frequency domain, which is why these methods are popular for process loops.

  • First order: AR = K/√(1 + ω²τ²), φ = −tan⁻¹ωτ.
  • Dead time: AR = 1, φ = −ωθ.
  • Bode: unstable if AR(ω_co) > 1.
  • GM = 1/AR(ω_co); PM = 180° + φ(AR = 1).

Example: An open loop 4/((s + 1)(0.5s + 1)(0.2s + 1)) has its phase crossover at ω = 4.12 rad/min. Find the gain margin.

  1. |D(jω)| = √(1 + 17) × √(1 + 4.25) × √(1 + 0.68) = 4.243 × 2.291 × 1.296 = 12.6.
  2. AR = 4/12.6 = 0.317.
  3. GM = 1/0.317 = 3.15.

Answer: GM = 3.15 (the gain could rise 3.15 times before instability).

Stability: all closed-loop roots in the left half-plane. Routh array tests it from coefficients; a zero row gives the ultimate gain and frequency. Root locus shows the roots moving with K_c. Frequency response: AR and phase; Bode diagram and Nyquist plot. Bode criterion: AR < 1 at phase crossover; gain margin and phase margin; Nyquist criterion for general cases. Dead time adds phase lag −ωθ.

  • Increasing gain moves roots toward instability.
  • Dead time limits the achievable gain.
  • GM 1.7–2 and PM 30–45° are typical targets.
  • Nyquist handles open-loop unstable systems.

Key formulas

  • AR = |G(jω)|, φ = ∠G(jω)
  • GM = 1/AR(ω_co)
  • PM = 180° + φ(ω_g)
  • K_u = a₂a₁/a₃ − 1 (three lags)

Practice questions

A system is stable if all roots of the characteristic equation

  1. are real
  2. have negative real parts
  3. are positive
  4. are zero
Show the answer

Answer: B. have negative real parts Left half-plane.

In a Routh array, two sign changes in the first column mean

  1. stable
  2. two RHP roots
  3. two zeros
  4. marginal stability
Show the answer

Answer: B. two RHP roots Count of RHP roots.

The phase lag of a dead time θ at frequency ω is

  1. ωθ
  2. tan⁻¹(ωθ)
  3. 90°
  4. 0
Show the answer

Answer: A. ωθ Linear in ω.

By the Bode criterion the loop is unstable if at the phase crossover AR is

  1. < 1
  2. > 1
  3. = 0
  4. negative
Show the answer

Answer: B. > 1 Signal grows round the loop.

A gain margin of 2 means the gain can be

  1. halved
  2. doubled before instability
  3. reduced to zero
  4. increased tenfold
Show the answer

Answer: B. doubled before instability Definition.

Theory questions

State Routh’s criterion and use it to find the range of K_c for stability of G_OL = K_c/((s + 1)(2s + 1)(3s + 1)). (6 marks)

Show the model answer

Characteristic: 6s³ + 11s² + 6s + 1 + K_c = 0. Routh: s³: 6, 6; s²: 11, 1 + K_c; s¹: (66 − 6(1 + K_c))/11; s⁰: 1 + K_c. Stability needs 66 − 6 − 6K_c > 0 ⇒ K_c < 10, and 1 + K_c > 0 ⇒ K_c > −1. So −1 < K_c < 10; K_u = 10. Auxiliary equation 11s² + 11 = 0 ⇒ ω_u = 1 rad/time, P_u = 2π.

Explain the Bode stability criterion, gain margin and phase margin. (5 marks)

Show the model answer

For an open loop with no RHP poles, the closed loop is unstable if AR > 1 at the phase-crossover frequency ω_co (φ = −180°): a sinusoid travelling round the loop is inverted by the comparator (another −180°), returns in phase and amplified, so oscillations grow. Gain margin GM = 1/AR(ω_co): the factor by which the gain can increase before instability (design 1.7–2). Phase margin PM = 180° + φ at the gain-crossover frequency where AR = 1: additional phase lag (e.g. extra dead time) the loop can tolerate (design 30–45°). Both protect against model uncertainty.

Controller tuning

Topics: Ziegler–Nichols tuning; Cohen–Coon method; IMC tuning.

Ziegler–Nichols, Cohen–Coon and IMC

Tuning chooses K_c, τ_I and τ_D for an acceptable response: stable with margin, fast enough, little overshoot, and not too sensitive to noise and model error. Criteria include quarter decay ratio and minimising integral errors (IAE, ISE, ITAE).

Ziegler–Nichols closed-loop (continuous cycling) method: with P control only, raise K_c until the loop oscillates steadily; note K_u and P_u. Settings: P K_c = 0.5K_u; PI K_c = 0.45K_u, τ_I = P_u/1.2; PID K_c = 0.6K_u, τ_I = P_u/2, τ_D = P_u/8. They give roughly quarter-decay responses, which many plants find too aggressive; Tyreus–Luyben settings (K_c = K_u/3.2, τ_I = 2.2P_u) are gentler. Pushing a real plant to sustained oscillation can be unsafe; a relay (auto-tune) test finds K_u and P_u with a small controlled oscillation.

Process reaction curve: with the loop open, step the controller output and record the response; fit a first-order-plus-dead-time model K e^(−θs)/(τs + 1) (tangent at the inflection point, or 28.3 % and 63.2 % points). Ziegler–Nichols open-loop settings: P K_c = τ/(Kθ); PI 0.9τ/(Kθ), τ_I = 3.33θ; PID 1.2τ/(Kθ), τ_I = 2θ, τ_D = 0.5θ. Cohen–Coon settings correct these for larger θ/τ. Internal model control (IMC) tuning uses the model directly with one parameter, the desired closed-loop time constant λ: for a PI controller on a FOPDT process K_c = τ/(K(λ + θ)), τ_I = τ (SIMC limits τ_I to 4(λ + θ)); choosing λ ≈ θ gives a good balance of speed and robustness.

  • ZN closed loop: K_u, P_u ⇒ 0.6K_u, P_u/2, P_u/8 (PID).
  • Reaction curve: K, τ, θ.
  • Cohen–Coon: better for larger θ/τ.
  • IMC PI: K_c = τ/(K(λ + θ)), τ_I = τ.

Example: A loop cycles continuously at K_c = 10 with a period of 3 min. Find ZN PI and PID settings.

  1. PI: K_c = 0.45 × 10 = 4.5; τ_I = 3/1.2 = 2.5 min.
  2. PID: K_c = 0.6 × 10 = 6; τ_I = 3/2 = 1.5 min; τ_D = 3/8 = 0.375 min.

Answer: PI 4.5 / 2.5 min; PID 6 / 1.5 / 0.375 min.

Example: A reaction curve gives K = 2, τ = 10 min, θ = 2 min. Find ZN and IMC (λ = θ) PI settings.

  1. ZN: K_c = 0.9τ/(Kθ) = 0.9 × 10/(2 × 2) = 2.25; τ_I = 3.33 × 2 = 6.66 min.
  2. IMC: K_c = τ/(K(λ + θ)) = 10/(2 × 4) = 1.25; τ_I = τ = 10 min.

Answer: ZN 2.25 / 6.7 min (aggressive); IMC 1.25 / 10 min (smoother).

Tuning criteria: quarter decay, IAE/ISE/ITAE, robustness. Ziegler–Nichols closed loop from K_u and P_u (or relay auto-tuning); Tyreus–Luyben gentler. Reaction curve gives a FOPDT model; Ziegler–Nichols open loop and Cohen–Coon formulas. IMC tuning with one parameter λ (≈ θ for a good balance).

  • ZN settings are aggressive; detune if oscillatory.
  • Large θ/τ makes a process hard to control.
  • IMC λ trades speed against robustness.
  • Always fine-tune on the plant.

Key formulas

  • ZN PID: K_c = 0.6K_u, τ_I = P_u/2, τ_D = P_u/8
  • ZN open-loop PI: K_c = 0.9τ/(Kθ), τ_I = 3.33θ
  • IMC PI: K_c = τ/(K(λ + θ)), τ_I = τ

Practice questions

In the ZN closed-loop method, the PI integral time is

  1. P_u/2
  2. P_u/1.2
  3. P_u/8
  4. 2P_u
Show the answer

Answer: B. P_u/1.2 Table value.

The process reaction curve is obtained with the loop

  1. closed
  2. open
  3. at K_u
  4. in cascade
Show the answer

Answer: B. open Step in controller output.

ZN settings aim for a decay ratio of about

  1. 0
  2. 1/4
  3. 1/2
  4. 1
Show the answer

Answer: B. 1/4 Quarter decay.

In IMC tuning the single parameter is

  1. K_u
  2. λ
  3. P_u
  4. θ
Show the answer

Answer: B. λ Closed-loop time constant.

A relay test is used to find

  1. K_u and P_u safely
  2. valve C_v
  3. dead time only
  4. noise level
Show the answer

Answer: A. K_u and P_u safely Auto-tuning.

Theory questions

Describe the Ziegler–Nichols closed-loop tuning method and its limitations. (5 marks)

Show the model answer

Put the controller on P only (τ_I = ∞, τ_D = 0). Increase K_c in small steps, making small set-point changes, until the output oscillates with constant amplitude: K_u = that gain, P_u = oscillation period. Settings: P: 0.5K_u; PI: 0.45K_u, P_u/1.2; PID: 0.6K_u, P_u/2, P_u/8. Limitations: driving a plant to the stability limit may be unsafe and upsets production; quarter-decay response is oscillatory and not robust; not applicable to processes that do not oscillate (first/second order without dead time); results depend on operating point. Relay auto-tuning and gentler rules (Tyreus–Luyben) address these.

Explain the process reaction curve method with the Cohen–Coon settings. (5 marks)

Show the model answer

Open the loop (manual), make a step change M in controller output and record the measured variable. Fit a FOPDT model: K = Δy/M; draw the tangent at the inflection point: its intercept with the initial value gives θ, and the time from there to where it meets the final value gives τ (or use the 28.3 % and 63.2 % times: τ = 1.5(t₆₃ − t₂₈), θ = t₆₃ − τ). Cohen–Coon: P: K_c = (τ/Kθ)(1 + θ/3τ); PI: K_c = (τ/Kθ)(0.9 + θ/12τ), τ_I = θ(30 + 3θ/τ)/(9 + 20θ/τ); PID: K_c = (τ/Kθ)(4/3 + θ/4τ), τ_I = θ(32 + 6θ/τ)/(13 + 8θ/τ), τ_D = 4θ/(11 + 2θ/τ).

Advanced process control

Topics: Introduction to multivariable control; cascade control; ratio control; split-range control; feedforward control; inferential control; adaptive control; distributed control systems; dead-time compensation.

Enhancements to single-loop feedback

Cascade control uses two controllers: the primary (master) controller of the main variable sets the set point of a secondary (slave) controller on a faster inner variable. Example: reactor temperature (primary) cascaded to jacket-coolant temperature or flow (secondary). Disturbances entering the inner loop (coolant pressure or temperature changes) are corrected before they upset the primary variable; the inner loop must be several times faster than the outer.

Ratio control keeps one flow in a fixed ratio to another, wild flow: fuel-to-air in burners, reflux ratio, reactant ratios, blending. Split-range control lets one controller output drive two valves over different parts of its range: for example 0–50 % opens a cooling valve and 50–100 % a heating valve on a reactor jacket, or two valves in parallel to extend rangeability. Feedforward control measures a major disturbance and acts before it affects the output, using G_f = −G_d/G_p; it is combined with feedback to remove residual error.

Inferential control controls a variable that cannot be measured quickly (composition) through one that can (a tray temperature, pressure-compensated) or through a soft sensor model. Adaptive control adjusts controller parameters as the process changes (gain scheduling, self-tuning regulators). Dead-time compensation (the Smith predictor) uses a model to predict the effect of control action before the delay, letting a higher gain be used. Multivariable processes have interactions between loops (e.g. top and bottom composition in a column); the relative gain array helps choose pairings, decouplers reduce interaction, and model predictive control (MPC) handles many inputs, outputs and constraints together. A distributed control system (DCS) implements all these in networked controllers with operator stations, historians and alarm management; PLCs handle discrete logic and safety interlocks.

  • Cascade: inner loop faster, rejects its disturbances.
  • Ratio: flow B = r × flow A.
  • Split range: one output, two valves.
  • Feedforward: G_f = −G_d/G_p, with feedback trim.
  • Smith predictor for dead time; MPC for multivariable.

Example: A burner needs 17 kg air per kg fuel with 10 % excess air. Fuel flow (wild) is 120 kg/h. What air set point does the ratio controller send?

  1. Ratio = 17 × 1.10 = 18.7 kg/kg.
  2. Air set point = 18.7 × 120 = 2244 kg/h.

Answer: 2244 kg/h of air.

Example: A feedforward controller is needed for G_p = 2e^(−s)/(5s + 1) and G_d = 1.5e^(−2s)/(3s + 1). Find G_f.

  1. G_f = −G_d/G_p = −(1.5/2)(5s + 1)/(3s + 1) e^(−(2 − 1)s).
  2. = −0.75(5s + 1)/(3s + 1) e^(−s).

Answer: Static gain −0.75 with a lead–lag 5/3 and a 1 min delay (realisable since the disturbance is slower).

Cascade (fast inner loop corrects its own disturbances), ratio (fixed flow ratios), split range (one output, two valves), feedforward (G_f = −G_d/G_p plus feedback), inferential (soft sensors), adaptive (gain scheduling, self-tuning), dead-time compensation (Smith predictor), multivariable control (interaction, relative gain array, decoupling, MPC), DCS and PLC implementation.

  • Inner loop should be 3–5 times faster than the outer.
  • Feedforward needs a measurable disturbance and a model.
  • MPC handles constraints explicitly.
  • Relative gain near 1 suggests a good pairing.

Key formulas

  • G_f = −G_d/G_p
  • Ratio: F_B,sp = r F_A
  • Smith predictor removes e^(−θs) from the characteristic equation

Practice questions

In cascade control the secondary loop should be

  1. slower than the primary
  2. faster than the primary
  3. open
  4. manual
Show the answer

Answer: B. faster than the primary Inner loop fast.

Fuel–air control in a furnace is an example of

  1. split range
  2. ratio control
  3. inferential control
  4. adaptive control
Show the answer

Answer: B. ratio control Fixed ratio.

A Smith predictor compensates for

  1. noise
  2. dead time
  3. valve stiction
  4. offset
Show the answer

Answer: B. dead time Delay.

Feedforward control acts on

  1. the error
  2. a measured disturbance
  3. the set point only
  4. the integral of error
Show the answer

Answer: B. a measured disturbance Before the output changes.

Controlling column composition through a tray temperature is

  1. inferential control
  2. ratio control
  3. split range
  4. on–off control
Show the answer

Answer: A. inferential control Indirect measurement.

Theory questions

Explain cascade control with an example and its advantages. (5 marks)

Show the model answer

Two controllers in series: the primary measures the main controlled variable and its output becomes the set point of the secondary, which measures an intermediate variable and drives the valve. Example: reactor temperature TC (primary) → jacket-outlet temperature TC (secondary) → cooling-water valve. Advantages: disturbances entering the secondary loop (cooling-water temperature or pressure) are corrected quickly before they affect the reactor; the secondary loop linearises and speeds up the effective process seen by the primary; valve problems are compensated. Requirements: the secondary must respond faster than the primary and be measurable; tune the secondary first, then the primary.

Describe feedforward control and why it is combined with feedback. (5 marks)

Show the model answer

Feedforward measures a disturbance D and adjusts the manipulated variable so that its effect cancels the disturbance before the output changes: from Y = G_dD + G_pG_fD = 0, G_f = −G_d/G_p (a gain, lead–lag and possibly delay). Example: steam to a heat exchanger adjusted from measured process flow. It cannot correct unmeasured disturbances or model errors, and perfect cancellation may need non-realisable prediction. Feedback is therefore added: the feedback controller trims the remaining error, while feedforward handles the major measured disturbance quickly.

Notes, examples and questions were written for this studio; the prescribed books are listed in the studio for reference.

Frequently asked questions

What is the Process Dynamics & Control Studio?

Model it, close the loop, tune it. Process models and Laplace transforms, interacting tanks, block diagrams, P/PI/PID control, control valves, Routh and Bode stability, Ziegler–Nichols, Cohen–Coon and IMC tuning, cascade, ratio and feedforward from the ASTU CHE241501 syllabus.

What can I do in the Process Dynamics & Control Studio?

Transfer functions, tanks and dynamics; Feedback loops, offset and PID modes; Routh, Bode, gain and phase margins; ZN, Cohen–Coon, IMC tuning and cascade.

Is the Process Dynamics & Control Studio free?

Yes. It is free and runs in your browser with no sign-up.

Open the interactive studio →