Process dynamics and transfer functions

In short

Process dynamics is the study of how a process variable (level, temperature, composition, pressure) changes with time after an input changes, and a transfer function is the Laplace-domain ratio of output response to input change for a linear system. It converts the differential equations of a mass or energy balance into algebra, so a tank, heater or reactor can be represented by a gain, a time constant and a dead time. Every control decision — controller tuning, loop stability, whether a loop can be cascaded — rests on this model. It carries heavy weight in GATE CH (the Process Dynamics and Control section), in university Instrumentation and Process Control papers, and in interviews at Reliance, IOCL, BPCL, HPCL, Tata Chemicals, UPL, Aditya Birla, L&T Technology Services, Honeywell, Yokogawa, Emerson and ABB for DCS, commissioning and APC roles.

Explained

Why dynamics, and the modelling route

• A steady-state balance gives the final value; dynamics gives the path and the time taken — needed for startup, grade change, surge and safety response.
• Route: write unsteady balance → linearise about steady state → take deviation variables → Laplace transform → rearrange to G(s) = Y(s)/U(s).
• Unsteady balance: accumulation = in − out + generation. For a tank: A dh/dt = Fin − Fout.
• Deviation (perturbation) variables: y = Y − Ȳ, u = U − Ū where bar is the steady state. This makes all initial conditions zero, which is why transfer functions have no constant term.
• Lumped parameter models (well-mixed tank) give ODEs; distributed models (plug-flow, long pipe) give PDEs and are usually approximated by dead time plus first order.
• A transfer function is valid only for a linear (or linearised) time-invariant system, for deviations about the chosen operating point.

Laplace transform tools used

• Definition: L{f(t)} = ∫₀^∞ f(t)e^{−st}dt; the derivative rule L{df/dt} = sF(s) − f(0) with f(0)=0 in deviation form.
• Standard inputs: step (U/s), impulse (1), ramp (R/s²), sinusoid (Aω/(s²+ω²)), rectangular pulse (A(1−e^{−t₀s})/s).
• Time-delay (dead time) theorem: L{f(t−θ)} = e^{−θs}F(s) — hence the e^{−θs} factor in FOPDT models.
• Final value theorem: y(∞) = lim_{s→0} sY(s) — the quick way to get the ultimate response; valid only for stable systems.
• Initial value theorem: y(0⁺) = lim_{s→∞} sY(s) — tells whether the response starts flat (strictly proper) or jumps.
• Inversion by partial fractions; repeated and complex roots give t·e^{−at} and damped sinusoid terms.

Linearisation

• Taylor expansion to first order: f(x) ≈ f(x̄) + (df/dx)|_{x̄}(x − x̄). Multivariable: add each partial derivative term.
• Classic cases: outlet flow through a valve Fout = k√h → linearised slope k/(2√h̄); Arrhenius rate k₀e^{−E/RT} → slope (E/RT̄²)·k̄; vapour pressure, enthalpy of a stream.
• The linear model is accurate only near x̄ — a large step away from the operating point gives a visible error, which is why loop tuning is re-checked at each load.
• The resistance of a linearised outlet, R = dh/dF = 2√h̄/k, and capacitance C = A (tank area) give τ = RC — the electrical analogy examiners like.

First-order systems

• Standard form: G(s) = Kp/(τs+1). Kp = process (steady-state) gain = Δoutput/Δinput in engineering units; τ = time constant (time).
• Step response of size M: y(t) = KpM(1 − e^{−t/τ}). At t = τ, 63.2% complete; 2τ → 86.5%; 3τ → 95%; 4τ → 98.2%; 5τ → 99.3% (treated as settled).
• Initial slope = KpM/τ — drawing the tangent at t = 0 and reading where it cuts the final value gives τ graphically.
• Self-regulating (capacity with resistance): a tank with a valve on the outlet finds a new level by itself.
• Pure capacity / integrating process: G(s) = Kp/s (tank with a constant-speed outlet pump, or level with fixed outflow) — no steady state, output ramps for a step input. Level loops in drums and surge tanks are of this type.
• Examples: mercury thermometer in a bath, well-mixed isothermal tank, pressure in a vessel with a restriction, concentration in a CSTR with first-order reaction (τ = V/(F + kV)).

Dead time and the FOPDT model

• Dead time (transport lag, θ) is the delay before any response appears: θ = pipe volume / volumetric flow = L/v, or the travel time on a conveyor or along a dryer.
• FOPDT (first order plus dead time): G(s) = Kp e^{−θs}/(τs+1) — the industrial workhorse; three parameters identified from one open-loop step (bump) test.
• Identification: Kp = Δy_∞/Δu; θ from the flat portion; τ from the 63.2% point measured after the delay, or by the two-point method (τ = 1.5(t_63 − t_28)).
• θ/τ ratio decides controllability: < 0.1 easy, 0.1–0.5 normal, > 1 difficult — needs a model-based or dead-time compensator (Smith predictor).
• Padé / Taylor approximations of e^{−θs} (e.g. e^{−θs} ≈ (1 − θs/2)/(1 + θs/2)) are used where a rational transfer function is required.

Second-order and higher-order systems

• Standard form: G(s) = Kp/(τ²s² + 2ζτs + 1), with ζ the damping ratio and τ the characteristic time; natural frequency ωn = 1/τ.
• ζ > 1 overdamped (two real roots, sluggish, no overshoot); ζ = 1 critically damped; 0 < ζ < 1 underdamped (oscillatory with decaying amplitude); ζ = 0 undamped; ζ < 0 unstable.
• Underdamped step response descriptors: overshoot, decay ratio (= OS²), rise time, period of oscillation T, settling time.
• Two non-interacting first-order tanks in series: G = Kp/((τ₁s+1)(τ₂s+1)) — always overdamped (ζ ≥ 1); a second-order process is underdamped only with inertia or feedback (e.g. a controlled loop, a pneumatic valve with spring and mass, a manometer).
• Interacting tanks (the second tank's level pushes back on the first): the same two time constants appear but the effective response is slower, and τ_eff² = τ₁τ₂ with 2ζτ_eff = τ₁ + τ₂ + A₁R₂.
• n equal first-order lags in series behave like an FOPDT with apparent dead time — the basis of the "N+1 rule of thumb" used in model reduction.

Poles, zeros, stability and special responses

• Poles = roots of the denominator; each pole −1/τᵢ contributes e^{−t/τᵢ}. Zeros = roots of the numerator; they shape the response but not stability.
• Stability criterion: a linear system is stable if and only if all poles have negative real parts (lie in the left half of the s-plane). One pole on the imaginary axis → sustained oscillation; positive real part → divergence.
• Lead-lag element Kp(τ_a s+1)/(τs+1): if τ_a > τ, the response overshoots its final value; used in feedforward compensators.
• Inverse response: a positive (right-half-plane) zero, arising when two opposing effects act at different speeds — classic case the boiler drum level ("shrink and swell") and some exothermic reactor temperatures. It limits achievable control performance and cannot be removed by tuning.
• Frequency response: substituting s = jω gives amplitude ratio AR = |G(jω)| and phase φ = ∠G(jω); a first-order lag has AR = Kp/√(1+ω²τ²), φ = −tan⁻¹(ωτ), approaching −90°; dead time adds phase −ωθ with no attenuation — which is why dead time destabilises loops so effectively. Plotted as Bode (log–log AR, semi-log φ) or Nyquist diagrams.
• Block diagram algebra: series blocks multiply; parallel add; a negative feedback loop reduces to G/(1 + GH) (forward path over 1 + loop gain). The closed-loop characteristic equation is 1 + Gc Gv Gp Gm = 0.
• Routh–Hurwitz test and direct substitution (s = jω) give the ultimate gain Kcu and ultimate period Pu without factorising.

Indian codes and practice

• Instrument symbols and loop tags follow ISA-5.1 (adopted in Indian engineering practice) with P&ID conventions; functional safety of loops follows IEC 61508 / IEC 61511, adopted in India as the IS/IEC 61508 and 61511 series.
• Control valve sizing and inherent characteristic (linear, equal percentage, quick opening) follow IEC 60534 series; the valve and transmitter dynamics (typically a few seconds) are part of the loop model and are often lumped into τ or θ.
• Factory and process safety obligations that drive interlock and trip dynamics now sit under the Occupational Safety, Health and Working Conditions Code, 2020, in force from 21 November 2025 (it replaced the Factories Act, 1948); hazardous-process plant rules continue under the State rules framed under it, alongside the MSIHC Rules.
• Transmitter output standards used in loop models: 4–20 mA, 3–15 psig for pneumatics, with HART/Foundation Fieldbus/Profibus for digital loops.

Key points

• Transfer function - Laplace ratio Y(s)/U(s) of output to input in deviation variables, with zero initial conditions.
• Deviation variable - measured minus steady-state value; it removes initial conditions and constants from the model.
• Process gain Kp - ultimate change in output per unit change in input (units = output units / input units).
• Time constant τ - time for 63.2% of the total change in a first-order step response; τ = RC = capacity × resistance.
• Dead time θ - pure delay before any response; adds phase lag without reducing amplitude, so it is the main destabiliser.
• FOPDT - Kp e^{−θs}/(τs+1); three parameters from one step test; the model behind almost all industrial tuning rules.
• Integrating process - Kp/s; no self-regulation, output ramps for a step; typical of level in a drum with a fixed-speed pump.
• Linearisation - first-order Taylor expansion about the operating point; valid only nearby (e.g. √h and Arrhenius terms).
• Damping ratio ζ - sets the shape: >1 overdamped, =1 critical, 0<ζ<1 underdamped (overshoot), <0 unstable.
• Decay ratio = (overshoot)² - the quarter-amplitude damping used in Ziegler–Nichols design corresponds to DR = 0.25.
• Non-interacting tanks in series - time constants multiply as separate lags; the system can never overshoot a step.
• Interacting tanks - downstream level affects upstream outflow; response is slower than the non-interacting case.
• Poles - denominator roots; they alone decide stability (all must lie in the left half-plane).
• Right-half-plane zero - causes inverse response (drum-level shrink and swell); a fundamental performance limit.
• Final value theorem - y(∞) = lim_{s→0} sY(s); the fastest route to steady-state gain checks.
• Frequency response - AR and phase from s = jω; first-order lag tends to −90°, each extra lag another −90°.
• Block reduction - series multiply, parallel add, negative feedback → G/(1+GH); characteristic equation 1 + loop gain = 0.
• Settling - 3τ ≈ 95%, 5τ ≈ 99% complete; used in startup and batch cycle-time estimates.

Formulas and examples

Example 1 — first-order tank: τ, Kp and the response. A vertical cylindrical tank of diameter 1.5 m has outflow through a valve, Fout = k√h with k = 0.004 m^2.5/s. Steady state: Fin = 0.008 m³/s.

• Steady level: √h̄ = 0.008/0.004 = 2 → h̄ = 4 m.
• Area A = π(1.5)²/4 = 1.767 m².
• Resistance R = 2√h̄/k = 2(2)/0.004 = 1,000 s/m².
• τ = RA = 1,000 × 1.767 = 1,767 s (29.4 min); Kp = R = 1,000 m per m³/s.
• Inflow is stepped up by M = 0.0005 m³/s. New steady level rise = KpM = 1,000 × 0.0005 = 0.5 m → h(∞) = 4.5 m.
• Level after 30 min (1,800 s): h = 4 + 0.5(1 − e^{−1800/1767}) = 4 + 0.5(1 − e^{−1.0187}) = 4 + 0.5(1 − 0.361) = 4.32 m.
• Time to reach 4.4 m (80% of the change): t = −τ ln(0.2) = 1,767 × 1.609 = 2,844 s (47.4 min).

Example 2 — FOPDT identification from a bump test. A heat-exchanger outlet temperature is logged after the steam valve is opened 5% (controller in manual). Temperature was steady at 70 °C, starts to move 12 s after the bump, and settles at 82 °C. It reads 77.6 °C at 52 s.

• Kp = Δy/Δu = (82 − 70)/5 = 2.4 °C per % valve.
• Dead time θ = 12 s.
• 63.2% point: 70 + 0.632(12) = 77.58 °C ≈ 77.6 °C, reached at 52 s → τ = 52 − 12 = 40 s.
• Model: G(s) = 2.4 e^{−12s}/(40s + 1); θ/τ = 0.30 → a normal, PI-controllable loop.
• Check with the final value theorem for a 5% step: Y(s) = [2.4e^{−12s}/(40s+1)]·(5/s) → y(∞) = 2.4 × 5 = 12 °C rise, i.e. 82 °C. ✓

Example 3 — second-order underdamped response. A closed temperature loop responds to a 4 °C setpoint step as Y/Ysp = 1/(9s² + 2.4s + 1) (time in minutes).

• τ² = 9 → τ = 3 min; 2ζτ = 2.4 → ζ = 2.4/(2 × 3) = 0.40 (underdamped).
• Overshoot = exp(−π × 0.4/√(1−0.16)) = exp(−1.2566/0.9165) = exp(−1.371) = 0.254 → first peak = 4(1.254) = 5.02 °C.
• Decay ratio = 0.254² = 0.0645.
• Period T = 2π(3)/√0.84 = 18.85/0.9165 = 20.6 min; first peak at T/2 = 10.3 min.
• ωn = 1/3 = 0.333 rad/min; ζ < 1 confirms poles are complex: s = (−2.4 ± √(5.76 − 36))/18 = −0.133 ± 0.306j — negative real part, so stable.

Example 4 — two tanks, non-interacting versus interacting. Tank 1: A₁ = 2 m², R₁ = 50 s/m² → τ₁ = 100 s. Tank 2: A₂ = 3 m², R₂ = 40 s/m² → τ₂ = 120 s.

• Non-interacting: G = R₂/((100s+1)(120s+1)); τ_eff = √(100 × 120) = 109.5 s; 2ζτ = τ₁+τ₂ = 220 → ζ = 220/(2 × 109.5) = 1.005 — just overdamped, no overshoot.
• Interacting: 2ζτ = τ₁ + τ₂ + A₁R₂ = 100 + 120 + (2 × 40) = 300 s → ζ = 300/(2 × 109.5) = 1.37.
• Interaction raised ζ from 1.00 to 1.37, so the response is slower (more sluggish) though the final value is unchanged.

Example 5 — dead time versus lag in frequency terms. Compare G₁ = 1/(10s+1) and G₂ = e^{−10s} at ω = 0.1 rad/s.

• Lag: AR = 1/√(1 + (0.1 × 10)²) = 1/√2 = 0.707; φ = −tan⁻¹(1) = −45°.
• Dead time: AR = 1.0; φ = −ωθ = −0.1 × 10 = −1 rad = −57.3°.
• At ω = 0.5 rad/s: lag gives AR = 1/√26 = 0.196, φ = −78.7°; dead time gives AR = 1, φ = −5 rad = −286°.
• Lesson: the lag attenuates the signal as phase builds up, dead time does not — so dead time drives the loop towards −180° at full amplitude and sets the stability limit.

Where it is used

• Refinery and petrochemical DCS/APC work (Reliance Jamnagar, IOCL Panipat, BPCL Kochi, HPCL, HMEL): FOPDT models from step tests feed PID tuning and multivariable controllers for crude columns and FCC units.
• Boiler and power plant control (NTPC, BHEL-supplied units, captive plants): drum-level inverse response forces three-element control; superheater temperature is modelled as lags plus dead time.
• Fertiliser, chemicals and specialty plants (RCF, GSFC, Tata Chemicals, UPL, Deepak Nitrite, Aarti Industries): CSTR and column dynamics for grade changes, startup sequences and interlock response times.
• Pharma batch processing (Sun Pharma, Cipla, Dr Reddy's): reactor jacket temperature dynamics for cascade control and validated batch recipes under GMP.
• Automation vendors and engineering firms (Honeywell, Yokogawa, Emerson, ABB, Siemens, L&T, Tata Consulting Engineers): loop checking, commissioning, simulation and operator-training simulators built on these models.
• Job roles: instrumentation/control engineer, DCS and commissioning engineer, APC engineer, process simulation engineer (Aspen Dynamics, HYSYS Dynamics, MATLAB/Simulink), plus GATE-based PSU recruitment through IOCL, GAIL, ONGC and BPCL.

Interview questions

• Q: What is a transfer function and what assumptions does it carry? A: It is the Laplace-domain ratio of output to input in deviation variables; it assumes a linear (or linearised) time-invariant system with zero initial conditions, valid near the operating point.
• Q: Why are deviation variables used? A: They make all initial conditions zero so the transfer function contains only dynamics, and they describe exactly what a controller sees — change from the steady state.
• Q: What is the physical meaning of a time constant? A: The time for 63.2% of the total change in a first-order step response; it equals capacitance × resistance, so a bigger holdup or tighter restriction makes the process slower.
• Q: How do you get Kp, τ and θ from plant data? A: Put the controller in manual, make a step change in output, then read Kp = Δoutput/Δinput at steady state, θ from the flat initial portion, and τ as the time from the end of the delay to 63.2% of the change.
• Q: Why is dead time more harmful than an equal amount of lag? A: Dead time adds phase lag with no amplitude attenuation, so the loop reaches −180° phase at full gain and goes unstable at a much lower controller gain.
• Q: Can two non-interacting first-order tanks in series overshoot a step? A: No — the poles are real, so ζ ≥ 1 and the response is overdamped; overshoot needs complex poles from feedback, inertia or numerator dynamics.
• Q: What causes inverse response and give a plant example? A: A right-half-plane zero from two opposing effects of different speed; boiler drum level falls momentarily when cold feedwater increases, because the colder water collapses the steam bubbles (shrink), before rising as mass accumulates.
• Q: What does a pole at the origin mean for a level loop? A: The process is integrating (non-self-regulating), so any flow imbalance makes the level ramp until it floods or empties — it needs feedback to hold at all.
• Q: How is stability judged from a transfer function? A: Stable if and only if every pole of the closed-loop transfer function has a negative real part; Routh–Hurwitz or direct substitution s = jω gives the limiting gain.
• Q: Why is a mercury thermometer in a thermowell slower than the bare bulb? A: The well adds thermal capacitance and a resistance (air gap or fill), creating an extra lag in series, so the effective time constant grows.
• Q: What is the θ/τ ratio used for? A: As a controllability index — below about 0.1 the loop is easy, around 0.1–0.5 ordinary PI works, above 1 the loop needs model-based control or a dead-time compensator.
• Q: Where does linearisation break down in practice? A: Far from the design point — square-root valve flow, Arrhenius reaction rate and pH neutralisation change gain sharply, so a single set of tuning constants does not hold over the whole range.

Common mistakes

• Writing the transfer function from the absolute variables instead of deviation variables, leaving a constant term that cannot exist.
• Measuring τ from the instant of the step instead of from the end of the dead time, which inflates τ and ruins the tuning.
• Treating the 63.2% rule as applying to second-order or higher systems — it is a first-order result only.
• Assuming series lags can overshoot, or expecting a first-order process to oscillate.
• Mixing up the time constant of the process with the integral time of the controller, and mixing seconds with minutes in a model.
• Using the final value theorem on an unstable or integrating system, where the limit has no physical meaning.
• Confusing a positive zero (inverse response, a real performance limit) with a positive pole (instability), or thinking tighter tuning removes inverse response.
• Ignoring valve and transmitter dynamics, then blaming the process when the measured response is slower than the model.

Related topics

Process control and instrumentation · Controller tuning and PID design · Frequency response and stability analysis · Advanced control: cascade, feedforward and ratio · Chemical reaction engineering: CSTR dynamics · Laplace transforms and differential equations