7.3 Continuous Control Strategies: Feedback, Feedforward, Cascade, Ratio & Advanced PID

7.3 Continuous Control Strategies: Feedback, Feedforward, Cascade, Ratio & Advanced PID

In continuous process manufacturing—such as oil refining, chemical synthesis, power generation, and water treatment—the core objective of automation is maintaining process variables (temperature, pressure, flow, level, pH) at desired setpoints despite severe load disturbances, raw material variations, and dynamic equipment non-linearities.

While simple single-input, single-output (SISO) feedback control suffices for straightforward loops, complex thermal and chemical processes demand advanced regulatory control strategies. A Certified Automation Professional (CAP) must master the mathematical foundations of Proportional-Integral-Derivative (PID) algorithms, determine correct control action directions relative to valve failure modes, and engineer advanced architectures including cascade, feedforward, ratio, split-range, and selective override control.

---

PID Control Algorithm Formulations

Although the acronym "PID" is universally recognized, controller implementations vary significantly across Distributed Control Systems (DCS) and Programmable Logic Controllers (PLCs). The three primary mathematical forms are the Standard (ISA Non-Interactive), Parallel (Ideal/Independent), and Series (Interactive) forms.

1. Standard / ISA (Non-Interactive) Form

Adopted by the International Society of Automation, the Standard Form scales both the integral and derivative actions by the controller gain ($K_p$):

$$u(t) = K_p \left( e(t) + \frac{1}{T_i} \int_{0}^{t} e(\tau)\,d\tau + T_d \frac{de(t)}{dt} \right) + u_{bias}$$

In the Laplace domain:

$$G_c(s) = K_p \left( 1 + \frac{1}{T_i s} + T_d s \right)$$

• $K_p$: Controller Gain (dimensionless or $\%/\%$). Related to Proportional Band ($PB$) by: $$PB = \frac{100\%}{K_p} \iff K_p = \frac{100\%}{PB}$$ A wide proportional band ($PB = 200\%$) corresponds to low, gentle gain ($K_p = 0.5$); a narrow proportional band ($PB = 20\%$) corresponds to aggressive, high gain ($K_p = 5.0$).
• $T_i$: Integral Time (minutes/repeat or seconds/repeat). Represents the time required for integral action to duplicate the proportional response under a step error. The reciprocal is the Reset Rate ($K_i$ or $R = 1/T_i$, measured in repeats/minute). Warning: Increasing $T_i$ decreases integral action, making the controller slower; increasing $K_i$ accelerates integral action.
• $T_d$: Derivative Time (minutes or seconds). Represents the time interval by which derivative action anticipates future error based on current rate of change.

2. Parallel (Independent) Form

In the Parallel Form, the proportional, integral, and derivative parameters act completely independently without a common multiplier:

$$u(t) = K_p e(t) + K_i \int_{0}^{t} e(\tau)\,d\tau + K_d \frac{de(t)}{dt} + u_{bias}$$ $$G_c(s) = K_p + \frac{K_i}{s} + K_d s$$

Adjusting controller gain $K_p$ in a parallel algorithm alters only proportional response, leaving integral and derivative contributions unchanged. Conversions between Standard and Parallel forms:

$$K_{i,parallel} = \frac{K_{p,standard}}{T_{i,standard}} \qquad K_{d,parallel} = K_{p,standard} \times T_{d,standard}$$

3. Series (Interactive) Form

Originating from the physical pneumatics of analog controllers (where internal pneumatic bellows interacted dynamically), the Series Form represents an interactive transfer function:

$$G_c(s) = K_p' \left( \frac{T_i' s + 1}{T_i' s} \right) (T_d' s + 1) = K_p' \left( \frac{T_i' + T_d'}{T_i'} + \frac{1}{T_i' s} + T_d' s \right)$$

The apparent parameters ($K_p', T_i', T_d'$) interact to create effective parameters:

$$K_{p,eff} = K_p' \left( \frac{T_i' + T_d'}{T_i'} \right) \qquad T_{i,eff} = T_i' + T_d' \qquad T_{d,eff} = \frac{T_i' T_d'}{T_i' + T_d'}$$

Derivative Implementation: Error vs. Process Variable

In a theoretical PID equation, derivative action operates on control error: $e(t) = SP(t) - PV(t)$. If an operator makes an abrupt step change to the setpoint ($SP$):

$$\frac{de(t)}{dt} = \frac{d(SP)}{dt} - \frac{d(PV)}{dt}$$

Because the derivative of a step change is mathematically an infinite impulse ($\delta(t)$), differentiating error produces a massive spike in controller output known as Derivative Kick, slamming the control valve wide open or shut and causing severe mechanical damage.

Industrial Practice: Modern continuous controllers compute derivative action strictly on the rate of change of the Process Variable (Derivative on PV):

$$u_D(t) = -K_p T_d \frac{d(PV)}{dt}$$

Furthermore, to prevent high-frequency electrical measurement noise from causing valve "chatter", a first-order low-pass filter is applied to the derivative term:

$$G_D(s) = -\frac{K_p T_d s}{1 + \alpha T_d s} \quad (\text{where } \alpha \approx 0.1)$$

---

Action Direction: Direct Acting vs. Reverse Acting

A continuous feedback controller must enforce negative feedback. The direction of the controller action depends on two factors: the physical process response and the fail-safe orientation of the final control element.

$$e(t) = \pm (SP - PV)$$

• Reverse Acting (typical for heating loops with fail-closed steam valves and for level loops that manipulate an inlet valve): $$PV \uparrow \implies u(t) \downarrow$$ As the Process Variable increases above Setpoint, the controller output decreases.
• Direct Acting (typical for cooling, flare vent, pump bypass, and level loops that manipulate an outlet valve): $$PV \uparrow \implies u(t) \uparrow$$ As the Process Variable increases above Setpoint, the controller output increases.

Systematic Rule for Loop Action Determination

+-----------------------------------------------------------------------------+
|                 CONTROLLER ACTION DETERMINATION MATRIX                      |
+-----------------------------------------------------------------------------+
| Process Type     | Valve Failure Mode | Valve Action     | Controller Action|
+-----------------------------------------------------------------------------+
| Heating (Steam)  | Fail Closed (FC)   | Air-to-Open      | REVERSE Acting   |
| Heating (Steam)  | Fail Open (FO)     | Air-to-Close     | DIRECT Acting    |
| Cooling (Chilled)| Fail Closed (FC)   | Air-to-Open      | DIRECT Acting    |
| Cooling (Chilled)| Fail Open (FO)     | Air-to-Close     | REVERSE Acting   |
| Tank Level Pump  | Fail Closed (FC)   | Air-to-Open (Out)| DIRECT Acting    |
+-----------------------------------------------------------------------------+

Example Analysis: Consider a steam-heated reboiler. For safety, the steam valve is Fail Closed (FC / Air-to-Open); a $4\text{ mA}$ ($0\%$) output shuts off steam, and $20\text{ mA}$ ($100\%$) delivers full heat.

1. Assume temperature ($PV$) rises above setpoint ($SP$).
2. To restore balance, steam flow must be reduced (valve must move toward closed).
3. Since the valve is Air-to-Open, closing the valve requires reducing air pressure, meaning controller output must decrease ($CO \downarrow$).
4. Therefore, as $PV \uparrow$, $CO \downarrow$. The controller MUST be configured as Reverse Acting.

If the steam valve were Fail Open (Air-to-Close), closing the valve would require increasing controller output ($CO \uparrow$), mandating a Direct Acting controller.

---

Anti-Reset Windup Mechanisms

When a sustained process disturbance or manual intervention causes a large control error that persists over time, the integral term keeps integrating:

$$u_I(t) = \frac{K_p}{T_i} \int_{0}^{t} e(\tau)\,d\tau$$

If the controller output reaches its physical limit (typically $0\%$ or $100\%$, or $4\text{ mA}$ / $20\text{ mA}$), the control valve cannot open or close any further. However, the mathematical integrator continues accumulating error, "winding up" to hundreds or thousands of percent.

When the process variable finally returns toward setpoint, the controller output cannot immediately respond. It remains pinned at saturation until the error reverses sign and integrates long enough to "unwind" the accumulated charge. This results in massive, prolonged process overshoot.

  PV / Output
      ^
      |         Process Variable (PV)
      |               /---\                  (Massive Overshoot from Windup)
   SP |--------------/-----\----------------------------------
      |             /       \______________
      |
 100% |=============\                        (Output Saturated at 100%)
      |              \                      
   0% +----------------\--------------------------------------> Time
                        ^ Integrator begins unwinding here (too late!)

Industrial Anti-Reset Windup Solutions

1. Integrator Clamping (Conditional Integration): The algorithm freezes numerical integration whenever the controller output reaches $0\%$ or $100\%$ and the error sign tends to drive the output further into saturation.
2. External Reset Feedback (ERF): The controller uses actual valve position feedback (from a smart positioner) or an unselected signal in an override selector as the base for its internal filter calculation, preventing the integral calculation from ever exceeding physical valve boundaries.

---

Cascade Control Architecture

Cascade control improves regulatory performance when severe disturbances enter the manipulated variable path before reaching the primary process.

+-------------------------------------------------------------------------+
|                        CASCADE CONTROL ARCHITECTURE                     |
|                                                                         |
|           Primary (Master) Loop              Secondary (Slave) Loop     |
| +---------+  SP_master  +------------+   SP_slave   +-------------+     |
| | Setpoint|============>| Master PID |=============>|  Slave PID  |==+  |
| +---------+             | (Slow:Temp)|              |  (Fast:Flow)|  |  |
|                         +------------+              +-------------+  |  |
|                              ^                            ^          |  |
|                              | PV_master                  | PV_slave |  |
|                     +-------------------+        +---------------+   |  |
|                     |  Primary Process  |        | Secondary Dist|   |  |
|                     | (Chemical Reactor)|<=======| (Steam Header)|<==+  |
|                     +-------------------+        +---------------+ (Valve)
+-------------------------------------------------------------------------+

Core Engineering Principles of Cascade Control

1. Architecture: Two standard PID controllers are connected in series. The Primary (Outer / Master) controller measures the critical product variable (e.g., reactor temperature). The Secondary (Inner / Slave) controller measures an intermediate variable (e.g., jacket steam flow or jacket temperature). The output of the master controller directly writes the remote setpoint (RSP) of the slave controller.
2. Disturbance Isolation: If upstream steam header pressure drops, the slave flow loop immediately senses the reduction in flow and opens the control valve to compensate within seconds, long before the temperature of the massive chemical vessel deviates.
3. Speed of Response Requirement: For a cascade architecture to be dynamically stable, the inner (slave) loop must be significantly faster than the outer (master) loop: $$\tau_{inner} \le \frac{1}{3} \text{ to } \frac{1}{5} \tau_{outer}$$ If the inner loop is not at least 3 to 5 times faster than the outer loop, the two controllers will dynamically interact, causing resonance, hunting, and loop instability.
4. Mandatory Tuning Sequence:
   • Step 1: Place the Master controller in Manual.
   • Step 2: Place the Slave controller in Auto with a local setpoint. Tune the slave loop for fast, tightly damped response (decay ratio $\approx 1/4$ or slight overshoot).
   • Step 3: Switch the Slave controller setpoint mode to Cascade (Remote SP).
   • Step 4: Place the Master controller in Auto and tune the master loop. Never attempt to tune the master loop while the slave loop is un-tuned or operating in manual.

---

Feedforward Control: Dynamic Lead-Lag Compensation

While feedback control is inherently reactive (it cannot initiate corrective action until an error has already manifested in the process variable), Feedforward Control is proactive. It measures a major incoming load disturbance directly, calculates the exact corrective action required, and adjusts the manipulated variable before the process variable deviates.

+-------------------------------------------------------------------------+
|                        COMBINED FEEDFORWARD-FEEDBACK                    |
|                                                                         |
|                          Measured Disturbance: D(s)                     |
|                          (e.g., Cold Feed Flow Rate)                    |
|                                  /        \                             |
|                                 /          \                            |
|          Feedforward           v            v                           |
|       +----------------------------+     +-------------------+          |
|       | Dynamic Compensator G_ff(s)|     | Disturbance Path  |          |
|       +----------------------------+     |       G_d(s)      |          |
|                      |                   +-------------------+          |
|                      v                             |                    |
|                      (+) <--- Manipulated          v                    |
|                       |       Variable: u(t)     Process                |
|   +------------+      |    +------------+        G_p(s)                 |
|   | Feedback   |======+===>|   Valve    |==========>(+)===> PV: Y(s)    |
|   | Controller |           +------------+                ^              |
|   +------------+                                         |              |
|         ^                                                |              |
|         +------------------------------------------------+              |
+-------------------------------------------------------------------------+

Mathematical Derivation of Dynamic Feedforward

The total process output $Y(s)$ governed by manipulated input $U(s)$ and disturbance $D(s)$ is:

$$Y(s) = G_p(s) U(s) + G_d(s) D(s)$$

To achieve perfect disturbance rejection, we require $Y(s) = 0$ for any disturbance variation $D(s)$:

$$0 = G_p(s) U(s) + G_d(s) D(s) \implies U(s) = -\frac{G_d(s)}{G_p(s)} D(s)$$

Thus, the ideal feedforward transfer function is:

$$G_{ff}(s) = -\frac{G_d(s)}{G_p(s)}$$

Modeling the disturbance path and process path as First-Order Plus Dead-Time (FOPDT) systems:

$$G_d(s) = \frac{K_d e^{-\theta_d s}}{\tau_d s + 1} \qquad G_p(s) = \frac{K_p e^{-\theta_p s}}{\tau_p s + 1}$$ $$G_{ff}(s) = -\left( \frac{K_d}{K_p} \right) \left( \frac{\tau_p s + 1}{\tau_d s + 1} \right) e^{-(\theta_d - \theta_p)s}$$

The Lead-Lag Compensator Parameters

1. Static Feedforward Gain ($K_{ff}$): $$K_{ff} = -\frac{K_d}{K_p}$$ Determines the steady-state ratio between the disturbance change and valve displacement.
2. Lead Time Constant ($T_{lead} = \tau_p$): Compensates for the lag in the process response. If the process is sluggish compared to the disturbance ($\tau_p > \tau_d$), $T_{lead} > T_{lag}$, producing an initial kick to accelerate valve action.
3. Lag Time Constant ($T_{lag} = \tau_d$): Compensates for the lag in the disturbance path. If the disturbance reaches the output slowly ($\tau_d > \tau_p$), $T_{lag} > T_{lead}$, slowing down valve action to prevent premature overcompensation.
4. Dead-Time Realizability Constraint (Physical Causality): The delay term is $e^{-(\theta_d - \theta_p)s}$.
   • If disturbance dead time is greater than process dead time ($\theta_d \ge \theta_p$), the delay is physically realizable using a standard dead-time buffer: $\theta_{ff} = \theta_d - \theta_p$.
   • If process dead time exceeds disturbance dead time ($\theta_p > \theta_d$), the required term becomes $e^{+(\theta_p - \theta_d)s}$, which requires predicting the future. A controller cannot act before a disturbance occurs; therefore, perfect dynamic feedforward cancellation is physically impossible when process dead time exceeds disturbance dead time. The system can only implement static and lead-lag compensation, relying on feedback to clean up the inevitable transient error.

---

Ratio Control & Cross-Limiting Combustion Safety

Ratio Control maintains a predetermined mathematical relationship between two flowing streams ($F_A$ and $F_B$):

$$R = \frac{F_{controlled}}{F_{wild}} \iff SP_{controlled} = R \times F_{wild}$$

• Wild Stream ($F_{wild}$): An uncontrolled process stream (e.g., effluent from an upstream unit).
• Controlled Stream ($F_{controlled}$): An actively throttled stream adjusted to match the wild stream ratio.

Cross-Limiting (Lead-Lag) Fuel-Air Combustion Control

In industrial boilers, fired heaters, and furnaces, maintaining the correct fuel-to-air ratio is a critical life-safety requirement. If fuel flow exceeds stoichiometric combustion air, unburned hydrocarbons accumulate in the firebox, risking a catastrophic furnace explosion (deflagration). Excess air must be maintained at all times.

To prevent fuel-rich mixtures during dynamic firing rate transitions, industrial systems implement Cross-Limiting Control using High and Low Signal Selectors (Auctioneering):

+-------------------------------------------------------------------------+
|                    CROSS-LIMITING COMBUSTION CONTROL                    |
|                                                                         |
|                          Firing Rate Demand                             |
|                           /              \                              |
|                          /                \                             |
|                         v                  v                            |
|                  +-------------+    +-------------+                     |
|                  | Low Selector|    |High Selector|                     |
|                  |    (LS)     |    |    (HS)     |                     |
|                  +-------------+    +-------------+                     |
|                    ^         |        ^         |                       |
|      Actual Air    |         |        |         | Actual Fuel           |
|      Flow / Ratio -+         |        |         +- Flow * Ratio         |
|                              v        v                                 |
|                         Fuel Flow   Air Flow                            |
|                         Set Point   Set Point                           |
+-------------------------------------------------------------------------+

Operational Dynamics

• When Firing Rate Increases (Load Demand $\uparrow$):
  • The High Selector for combustion air immediately passes the increased firing demand, raising the Air Flow Setpoint instantly.
  • The Low Selector for fuel compares the increased firing demand against the actual measured airflow. The fuel setpoint cannot increase until the air flow has physically increased.
  • Result: AIR LEADS FUEL ON FIRING INCREASES.
• When Firing Rate Decreases (Load Demand $\downarrow$):
  • The Low Selector for fuel immediately passes the decreased firing demand, cutting the Fuel Flow Setpoint instantly.
  • The High Selector for air compares the decreased firing demand against the actual measured fuel flow. The air setpoint cannot decrease until the fuel flow has physically dropped.
  • Result: FUEL LEADS AIR ON FIRING DECREASES.

Through this cross-limiting selector logic, the burner is physically guaranteed to never operate in a fuel-rich condition during dynamic load swings.

---

Split-Range & Override (Selective) Control

Split-Range Control

In Split-Range Control, a single controller output ($0-100\%$) coordinates two or more final control elements operating across complementary spans:

  Valve Position (% Open)
 100% |  Cooling Valve (Air-to-Open)        Heating Valve (Air-to-Open)
      |   \                                       /
      |    \                                     /
      |     \                                   /
      |      \                                 /
   0% +-------\---------------|---------------/-------------> Controller Output
      0%      45%            50%             55%           100%
                              ^ Deadband (Both Shut)

• $0\% - 45\%$ Controller Output: Cooling valve throttles from $100\%$ open down to $0\%$ open. Heating valve is shut.
• $45\% - 55\%$ Controller Output (Deadband): Both cooling and heating valves remain fully closed to prevent wasteful simultaneous heating and cooling.
• $55\% - 100\%$ Controller Output: Heating valve throttles from $0\%$ open up to $100\%$ open. Cooling valve is shut.

Override / Selective Control

Override (or auctioneering) control protects equipment boundaries when process conditions approach structural or safety limits. Multiple PID controllers share a single final control element via a High Signal Selector (HS) or Low Signal Selector (LS).

Example: Gas Pipeline Compressor Protection:

• Controller 1 (Normal): Regulates pipeline suction pressure (Standard PID).
• Controller 2 (Override): Monitors compressor motor amperage. If amperage approaches maximum motor nameplate rating, Controller 2's output drops.
• A Low Selector routes the lower of the two controller outputs to the suction throttle valve. Under normal conditions, Controller 1 modulates the valve. If motor current spikes, Controller 2 takes over control seamlessly, overriding pressure control to prevent tripping the motor circuit breaker.
• Anti-Windup Mandate: The unselected controller's integrator must track the selected output (External Reset Feedback); otherwise, its integrator will wind up to $100\%$, creating a severe delay when an emergency override condition occurs.

---

Realistic Industrial Scenario: Exothermic Polymerization Reactor Control

A chemical manufacturing plant operates an exothermic continuous stirred-tank reactor (CSTR) producing specialty acrylic resins. The process exhibits rapid runaway kinetics if temperature exceeds $85^\circ\text{C}$.

Integrated Control Strategy Implementation

1. Primary Temperature / Jacket Cascade:
   • Primary Master Controller (TIC-101) measures internal reactor temperature ($PV_1$).
   • Secondary Slave Controller (TIC-102) measures reactor jacket coolant temperature ($PV_2$).
   • Slave loop time constant $\tau_{slave} = 45\text{ seconds}$; Master vessel time constant $\tau_{master} = 320\text{ seconds}$. Because $\tau_{master} / \tau_{slave} = 7.1 > 5.0$, the cascade architecture satisfies the dynamic speed requirement.
2. Feedforward Load Compensation:
   • A Coriolis flowmeter measures incoming cold monomer feed rate ($F_{feed}$). Monomer feed rate swings of $\pm 20\%$ occur frequently due to upstream distillation column cycling.
   • Disturbance gain $K_d = -0.45^\circ\text{C} / (\text{kg/min})$; Process gain $K_p = +1.20^\circ\text{C} / \% CO$.
   • Static Feedforward Gain: $K_{ff} = -(-0.45 / 1.20) = +0.375\% CO / (\text{kg/min})$.
   • Step tests gave a process-path time constant of 320 s and a disturbance-path time constant of 90 s, so the lead-lag compensator was configured with $T_{lead} = 320\text{ s}$ and $T_{lag} = 90\text{ s}$. The feedforward output is summed into the master controller output, which is the jacket temperature setpoint of the slave loop.
3. Split-Range Temperature Control:
   • Secondary slave controller output ($0-100\%$) manipulates two control valves via split-ranging: $0-48\%$ throttles chilled glycol cooling water; $52-100\%$ throttles low-pressure steam for reactor startup, with a $4\%$ deadband.

During commissioning, upstream feed flow abruptly increased by $15\text{ kg/min}$. The feedforward compensator added $15 \times 0.375 \approx 5.6\%$ to the master output, raising the jacket temperature setpoint to offset the cooling effect of the extra cold feed, and the reactor temperature excursion stayed within $\pm 0.3^\circ\text{C}$.

---

Common Exam Traps

1. Tuning Cascade Master Before Slave: Attempting to tune the primary (master) loop while the secondary (slave) loop is still in manual or poorly tuned is a major error. The inner slave loop alters the effective process dynamics seen by the master; therefore, the slave loop must always be tuned and placed in cascade mode first.
2. Confusing Reset Time ($T_i$) with Reset Rate ($K_i$): On the exam, verify units. If integral action is specified in minutes/repeat ($T_i$), a larger number means weaker, slower integral action. If specified in repeats/minute ($K_i$), a larger number means stronger, faster integral action.
3. Inverting Cross-Limiting Selectors: In boiler combustion cross-limiting control, connecting the air setpoint to a Low Selector and fuel setpoint to a High Selector creates a fatal hazard: fuel would lead air on firing increases, creating a fuel-rich explosive atmosphere in the firebox.
4. Neglecting Valve Fail-Safe Direction in Loop Action: Determining whether a controller is Direct or Reverse Acting cannot be deduced from process physics alone. You must know the valve failure position (Air-to-Open vs Air-to-Close). Reversing valve action without changing controller action converts negative feedback into positive feedback, driving the loop into immediate saturation.