18.2 PID Controller Tuning and Loop Stability

18.2 PID Controller Tuning and Loop Stability

In chemical manufacturing, improper controller tuning leads to catastrophic consequences: unstable exothermic reactors can experience thermal runaway, distillation columns can flood or dump trays, and uncoordinated pressure loops can lift relief valves. Achieving robust stability requires an analytical understanding of how controller parameters ($K_c, \tau_I, \tau_D$) interact with process inertia and dead time.

On the NCEES PE Chemical Exam, stability analysis and tuning questions require finding the ultimate gain ($K_u$) and period ($P_u$), constructing Routh arrays, evaluating Gain Margin ($GM$) and Phase Margin ($PM$) on Bode plots, and applying Ziegler-Nichols, Cohen-Coon, and Internal Model Control (IMC) tuning rules.

---

1. Closed-Loop Stability & Pole Placement

The Characteristic Equation

The stability of any linear feedback control loop is dictated by the denominator of its closed-loop transfer function. Regardless of whether a setpoint change or load disturbance occurs, stability depends entirely on the roots of the characteristic equation:

$$1 + G_{OL}(s) = 0 \iff 1 + G_c(s) G_v(s) G_p(s) G_m(s) = 0$$

Where $G_{OL}(s)$ is the total open-loop transfer function. The roots of this equation are the closed-loop poles ($s = p_i = \sigma_i + j \omega_i$):

• Stable System: All poles reside strictly in the open Left-Half Plane (LHP) of the complex $s$-plane: $\text{Re}(p_i) = \sigma_i < 0$ for all $i$. In the time domain, transient terms decay exponentially ($e^{\sigma_i t} \to 0$).
• Marginally Stable System: One or more non-repeated conjugate poles lie exactly on the imaginary axis ($\sigma_i = 0, \omega_i \neq 0$). The process exhibits sustained, constant-amplitude oscillations at frequency $\omega_{co}$.
• Unstable System: Any pole lies in the Right-Half Plane (RHP): $\text{Re}(p_i) = \sigma_i > 0$. The transient response grows exponentially ($e^{\sigma_i t} \to \infty$), resulting in physical saturation or equipment failure.

                    IMAGINARY AXIS (j*omega)
                               ^
      UNSTABLE REGION          |          STABLE REGION
     (Roots Grow: e^+sigma*t)  |    (Roots Decay: e^-sigma*t)
                               |          X  p1 = -sigma + j*omega
                               |
      -------------------------+-------------------------> REAL AXIS (sigma)
                               |
                               |          X  p2 = -sigma - j*omega
                               |
                 MARGINALLY STABLE (sigma = 0)
                 Continuous Cycling at omega_co

Dynamic Damping and Decay Ratio

For second-order closed-loop poles ($s^2 + 2 \zeta \omega_n s + \omega_n^2 = 0$ with $0 < \zeta < 1$):

• Damping Ratio ($\zeta$): Controls oscillation damping. Low $\zeta$ yields sharp oscillations.
• Decay Ratio ($DR$): The ratio of the amplitude of the second peak to that of the first peak: $$DR = \exp\left( -\frac{2 \pi \zeta}{\sqrt{1 - \zeta^2}} \right) = (OS)^2$$ The classic quarter-amplitude decay ratio ($DR = 1/4 = 0.25$) corresponds to $\zeta \approx 0.36$. While historically popular for rapid recovery, it is considered too oscillatory for modern, tightly integrated chemical processes where $\zeta \approx 0.5 - 0.7$ ($DR \le 0.05$) is preferred.

---

2. Algebraic Stability: The Routh-Hurwitz Criterion

The Routh-Hurwitz criterion tests whether any roots of a polynomial characteristic equation have positive real parts without requiring numerical root finding.

$$a_n s^n + a_{n-1} s^{n-1} + a_{n-2} s^{n-2} + \dots + a_1 s + a_0 = 0$$

Necessary Condition (Stodola's Rule)

All polynomial coefficients $a_n, a_{n-1}, \dots, a_0$ must be strictly non-zero and share the same mathematical sign (conventionally all positive, $a_i > 0$). If any coefficient is zero or negative, the system is automatically unstable.

Constructing the Routh Array

Arrange the polynomial coefficients into a Routh array:

Where the array elements are calculated via cross-multiplication:

$$b_1 = \frac{a_{n-1} a_{n-2} - a_n a_{n-3}}{a_{n-1}}, \quad b_2 = \frac{a_{n-1} a_{n-4} - a_n a_{n-5}}{a_{n-1}}$$ $$c_1 = \frac{b_1 a_{n-3} - a_{n-1} b_2}{b_1}, \quad c_2 = \frac{b_1 a_{n-5} - a_{n-1} b_3}{b_1}$$

Stability Theorem

The system is stable if and only if all elements in the first column of the Routh array are strictly positive ($a_n, a_{n-1}, b_1, c_1, \dots > 0$). The number of sign changes in the first column equals the number of unstable roots with positive real parts (in the RHP).

To find the ultimate controller gain $K_u$, set the row element containing $K_c$ to zero ($b_1 = 0$ or $c_1 = 0$). The auxiliary equation formed by the preceding row yields the sustained oscillation frequency $\omega_{co}$.

---

3. Frequency Response & Bode Stability Criteria

Frequency response evaluates loop behavior when subjected to sinusoidal inputs by substituting $s = j\omega$ into the open-loop transfer function: $G_{OL}(j\omega) = |G_{OL}(j\omega)| e^{j \phi(\omega)}$.

$$\text{Amplitude Ratio (AR):} \quad AR_{OL}(\omega) = |G_{OL}(j\omega)| = |G_c| \cdot |G_v| \cdot |G_p| \cdot |G_m|$$ $$\text{Total Phase Angle (}\phi\text{):} \quad \phi_{OL}(\omega) = \angle G_{OL}(j\omega) = \angle G_c + \angle G_v + \angle G_p + \angle G_m$$

   BODE PLOT: GAIN AND PHASE MARGIN DEFINITIONS
   Log AR
     ^
 1.0 +==================\============================ (0 dB Line)
     |                   \     Gain Crossover (omega_g: AR = 1)
     |                    \ 
     |                     *-----------------+ 
     |                      \                | Gain Margin (GM = 1 / AR_co)
     |                       \               v
     +------------------------*----------------------> Frequency (omega)
                               \ Phase Crossover (omega_co: phi = -180°)
   Phase (deg)                  \ 
     |                           \ 
-135°|                            \    +------+ Phase Margin
     |                             \   |      | (PM = 180° + phi_g)
-180°+------------------------------*--+------v------
     |                              |
     v                              v omega_co

Characteristic Frequencies

1. Phase Crossover Frequency ($\omega_{co}$): The frequency where total open-loop phase lag reaches $-180^\circ$ ($-\pi\text{ radians}$): $$\phi_{OL}(\omega_{co}) = \angle G_{OL}(j\omega_{co}) = -180^\circ$$
2. Gain Crossover Frequency ($\omega_g$): The frequency where open-loop amplitude ratio is exactly unity ($0\text{ dB}$): $$AR_{OL}(\omega_g) = |G_{OL}(j\omega_g)| = 1.0$$

The Bode Stability Criterion

A closed-loop system is stable if and only if the open-loop amplitude ratio is strictly less than 1.0 at the phase crossover frequency:

$$AR_{OL}(\omega_{co}) < 1.0$$

If $AR_{OL}(\omega_{co}) = 1.0$, the loop is marginally stable and oscillates continuously. If $AR_{OL}(\omega_{co}) > 1.0$, signals circulating around the loop grow with each cycle, causing severe instability.

Robustness Metrics: Gain Margin and Phase Margin

To safeguard against modeling errors and plant aging, chemical engineering loops are tuned to maintain strict stability margins:

$$\text{Gain Margin (GM):} \quad GM = \frac{1}{|G_{OL}(j\omega_{co})|} \ge 1.7 \text{ to } 2.0$$ $$GM_{\text{dB}} = 20 \log_{10}(GM) = -20 \log_{10} |G_{OL}(j\omega_{co})| \ge 4.6 \text{ to } 6.0\text{ dB}$$ $$\text{Phase Margin (PM):} \quad PM = 180^\circ + \phi_{OL}(\omega_g) \ge 30^\circ \text{ to } 45^\circ$$

Phase Angles for Elementary Dynamic Elements

• Dead Time ($e^{-\theta s}$): $AR = 1.0$, $\phi = -\theta \omega\text{ radians} = -\frac{180^\circ}{\pi} \theta \omega$. Introduces unbounded phase lag without gain attenuation, making dead time the primary limiter of loop performance!
• First-Order Lag ($1/(\tau s + 1)$): $AR = 1/\sqrt{1 + \omega^2 \tau^2}$, $\phi = -\arctan(\omega \tau)$.
• Pure Integrator ($1/s$): $AR = 1/\omega$, $\phi = -90^\circ$.
• Ideal Derivative ($s$): $AR = \omega$, $\phi = +90^\circ$.

---

4. Empirical Controller Tuning Methodologies

1. Ziegler-Nichols Closed-Loop (Ultimate Gain) Method

The Ziegler-Nichols (Z-N) closed-loop technique (also known as the continuous cycling method) is performed with the controller installed in automatic mode:

1. Deactivate integral action (set $\tau_I \to \infty$, or reset rate $= 0$) and derivative action (set $\tau_D = 0$).
2. Introduce a small setpoint disturbance.
3. Incrementally increase controller gain $K_c$ until the process variable exhibits sustained, constant-amplitude oscillations.
4. Record this critical gain as the Ultimate Gain ($K_u$) and measure the period of oscillation as the Ultimate Period ($P_u$): $$P_u = \frac{2 \pi}{\omega_{co}}$$

2. Tyreus-Luyben Tuning (Conservative Chemical Process Rules)

Ziegler-Nichols settings yield a quarter-amplitude decay ratio ($DR = 0.25$), which is often too aggressive for chemical reactors and distillation columns. The Tyreus-Luyben rules provide a more conservative, robust response:

• PI Control: $K_c = 0.31 \, K_u$, $\tau_I = 2.20 \, P_u$.
• PID Control: $K_c = 0.31 \, K_u$, $\tau_I = 2.20 \, P_u$, $\tau_D = P_u / 6.30$.

3. Open-Loop Step Testing: Ziegler-Nichols vs. Cohen-Coon

When closed-loop cycling is forbidden for safety reasons, the loop is decoupled to manual mode, and a step change in valve output $\Delta u$ is introduced to obtain the process reaction curve ($K_p, \tau_p, \theta$):

$$\text{Controllability Ratio:} \quad \chi = \frac{\theta}{\tau_p}$$

• If $\chi < 0.20$: Easy to control; standard Z-N open-loop rules work well.
• If $0.20 \le \chi \le 0.50$: Moderate difficulty.
• If $\chi > 0.50$: Dead-time dominated; classic Z-N tuning produces severe instability. Cohen-Coon or IMC rules must be applied.

4. Internal Model Control (IMC) Tuning

Internal Model Control parameterizes the controller in terms of the desired closed-loop time constant $\tau_c$ (a tuning knob where smaller $\tau_c$ gives faster response, while larger $\tau_c$ increases robustness):

$$K_c = \frac{1}{K_p} \left( \frac{\tau_p}{\tau_c + \theta} \right), \quad \tau_I = \tau_p$$

Typically, set $\tau_c = \theta$ for moderate performance, or $\tau_c = 3 \theta$ for high robustness. IMC produces smooth, monotonic setpoint tracking with zero overshoot.

---

5. Summary Comparison Table: PID Tuning Strategies

---

6. Comprehensive Worked Numerical Example

Problem Statement

A chemical reactor temperature control loop consists of three first-order non-interacting thermal capacitances in series, governed by the following open-loop transfer function ($G_v = 1.0, G_m = 1.0$):

$$G_{OL}(s) = \frac{2.0 \, K_c}{(s + 1)(2s + 1)(4s + 1)}$$

1. Routh-Hurwitz Stability: Formulate the closed-loop characteristic polynomial, construct the Routh array, and determine the range of controller gain $K_c$ for closed-loop stability.
2. Ultimate Parameters: Evaluate the ultimate gain $K_u$, phase crossover frequency $\omega_{co}$, and ultimate period $P_u$.
3. Bode Verification: Verify $\omega_{co}$ and $K_u$ using frequency response principles.
4. PID Tuning Calculations: Calculate the controller settings ($K_c, \tau_I, \tau_D$) using classic Ziegler-Nichols closed-loop rules, and compare them with Tyreus-Luyben settings.

---

Step 1: Characteristic Polynomial & Routh Array

The closed-loop characteristic equation is $1 + G_{OL}(s) = 0$:

$$1 + \frac{2.0 \, K_c}{(s + 1)(2s + 1)(4s + 1)} = 0$$ $$(s + 1)(2s + 1)(4s + 1) + 2.0 \, K_c = 0$$

Expand the product of terms:

$$(s + 1)(2s + 1) = 2s^2 + 3s + 1$$ $$(2s^2 + 3s + 1)(4s + 1) = 8s^3 + 2s^2 + 12s^2 + 3s + 4s + 1 = 8s^3 + 14s^2 + 7s + 1$$

Adding the controller gain yields the characteristic polynomial:

$$8s^3 + 14s^2 + 7s + (1 + 2 K_c) = 0$$

Construct the Routh array:

Compute $b_1$:

$$b_1 = \frac{(14)(7) - (8)(1 + 2 K_c)}{14} = \frac{98 - 8 - 16 K_c}{14} = \frac{90 - 16 K_c}{14}$$

For stability, all elements in the first column must be strictly positive:

1. $8 > 0$ and $14 > 0$ (satisfied)
2. $b_1 > 0 \implies 90 - 16 K_c > 0 \implies 16 K_c < 90 \implies K_c < 5.625$
3. $1 + 2 K_c > 0 \implies K_c > -0.500$

Thus, the range of controller gain for closed-loop stability is:

$$\mathbf{-0.500 < K_c < 5.625}$$

---

Step 2: Ultimate Gain and Period

The upper stability boundary defines the Ultimate Gain:

$$K_u = \mathbf{5.625}$$

At $K_c = K_u = 5.625$, the $s^1$ row vanishes ($b_1 = 0$). Form the auxiliary equation from the $s^2$ row:

$$14 s^2 + (1 + 2 K_u) = 0 \implies 14 s^2 + (1 + 2(5.625)) = 0$$ $$14 s^2 + 12.25 = 0 \implies s^2 = -\frac{12.25}{14} = -0.875$$ $$s = \pm j \sqrt{0.875} = \pm j \, 0.93541 \text{ rad/min}$$

The phase crossover frequency is:

$$\omega_{co} = \mathbf{0.9354 \text{ rad/min}}$$

The ultimate period of oscillation is:

$$P_u = \frac{2 \pi}{\omega_{co}} = \frac{2 \pi}{0.93541} = \mathbf{6.717 \text{ min}}$$

---

Step 3: Frequency Response (Bode) Verification

At phase crossover, the sum of individual phase angles must equal $-180^\circ$:

$$\phi(\omega) = -\arctan(\omega) - \arctan(2\omega) - \arctan(4\omega) = -180^\circ$$

Substitute $\omega = 0.93541\text{ rad/min}$:

• $-\arctan(0.93541) = -43.09^\circ$
• $-\arctan(2 \times 0.93541) = -\arctan(1.8708) = -61.88^\circ$
• $-\arctan(4 \times 0.93541) = -\arctan(3.7416) = -75.03^\circ$
• $\sum \phi = -43.09^\circ - 61.88^\circ - 75.03^\circ = -180.00^\circ$ (Verified!)

Compute the amplitude ratio of the process at $\omega_{co}$:

$$|G_p(j\omega_{co})| = \frac{2.0}{\sqrt{1 + \omega^2} \sqrt{1 + 4\omega^2} \sqrt{1 + 16\omega^2}}$$ $$1 + \omega^2 = 1 + 0.875 = 1.875 \implies \sqrt{1.875} = 1.3693$$ $$1 + 4\omega^2 = 1 + 3.500 = 4.500 \implies \sqrt{4.500} = 2.1213$$ $$1 + 16\omega^2 = 1 + 14.00 = 15.00 \implies \sqrt{15.00} = 3.8730$$ $$|G_p(j\omega_{co})| = \frac{2.0}{(1.3693)(2.1213)(3.8730)} = \frac{2.0}{11.250} = 0.17778$$

At marginal stability, $|G_{OL}| = K_u |G_p| = 1.0$:

$$K_u = \frac{1}{|G_p(j\omega_{co})|} = \frac{1}{0.17778} = \mathbf{5.625} \quad (\text{Exact Match!})$$

---

Step 4: PID Controller Settings Comparison

• Classic Ziegler-Nichols Closed-Loop PID Settings:

  $$K_c = 0.60 \, K_u = 0.60 \times 5.625 = \mathbf{3.375}$$ $$\tau_I = \frac{P_u}{2.0} = \frac{6.717\text{ min}}{2.0} = \mathbf{3.359\text{ min}}$$ $$\tau_D = \frac{P_u}{8.0} = \frac{6.717\text{ min}}{8.0} = \mathbf{0.840\text{ min}}$$

• Tyreus-Luyben Chemical Process Settings:

  $$K_c = 0.31 \, K_u = 0.31 \times 5.625 = \mathbf{1.744}$$ $$\tau_I = 2.20 \, P_u = 2.20 \times 6.717\text{ min} = \mathbf{14.777\text{ min}}$$ $$\tau_D = \frac{P_u}{6.30} = \frac{6.717\text{ min}}{6.30} = \mathbf{1.066\text{ min}}$$

(Notice that Tyreus-Luyben slashes the controller gain by nearly 50% and increases integral time by over 400%, suppressing oscillations and providing superior gain and phase margins).

---

7. Critical PE Exam Traps & Pitfalls