Answer: The difference lies in how the three terms (P, I, D) interact mathematically. In a Parallel PID form, each action is computed independently based on the error and summed at the end. In a Series (Interacting) PID form, the controller is modeled as a PI block in series with a PD block, meaning the derivative action affects the integral action and vice versa.

If you calculate tuning parameters using a parallel formula (like Ziegler-Nichols) and input them into a DCS that uses the series form (like old Foxboro, Fisher-Rosemount, or some Honeywell controllers) without converting them, the controller will not behave as expected. Specifically, the effective integral time will be longer and the effective gain will be different, potentially causing sluggish response or severe instability/oscillations.

Formulas for Conversion (Parallel to Series):

$$K_{c,series} = K_{c,parallel} \cdot f \quad T_{i,series} = T_{i,parallel} \cdot f \quad T_{d,series} = \frac{T_{d,parallel}}{f}$$

Where conversion factor: $$f = \frac{1 + \sqrt{1 - \frac{4 T_{d,parallel}}{T_{i,parallel}}}}{2}$$

Answer: When an operator changes the setpoint (SP) abruptly (a step change), the error $e(t) = SP(t) - PV(t)$ undergoes a mathematically instantaneous jump. Since the derivative term is proportional to the rate of change of error ($de/dt$), the derivative of a step change is an infinite spike. This causes the controller output to saturate instantly, causing a violent physical "kick" to the control valve or actuator, leading to wear, chattering, and system stress.

To solve this, industrial controllers implement Derivative on PV (PI-D). In this configuration, derivative action is calculated only on the negative rate of change of the Process Variable ($-d(PV)/dt$) rather than the error. Because the PV cannot change instantaneously due to process inertia, the derivative spike is completely eliminated. Another solution is applying a first-order low-pass filter to the derivative term (filter time constant $T_f = T_d / 10$) to damp high-frequency noise spikes.

Answer: Integral Windup occurs when a feedback loop has a sustained error that the controller cannot eliminate because the actuator (e.g., control valve) is fully open (100%) or closed (0%). The integral term continues to accumulate the error over time, causing the theoretical controller output value to grow far beyond the physical saturation limits (e.g. winding up to 1000%).

When the process variable finally crosses the setpoint, the controller cannot immediately react because the accumulated integral term must first "wind down". This results in massive overshoot, process oscillations, and delayed recovery, which can be extremely dangerous in chemical reactors or thermal systems.

Industrial Anti-Windup Solutions:

• Clamping (Limit Integration): Freeze the integral summation as soon as the controller output hits 0% or 100%, and only allow integration in the direction that moves the output away from saturation.
• Back-Calculation (Reset Feedback): Re-compute the integral term dynamically when saturation occurs, forcing the controller output to stay exactly at the actuator limit.

Answer: An Open-Loop Step Test is conducted by placing the controller in Manual Mode, introducing a step change in the Controller Output (CO) (e.g. opening a valve from 40% to 50%), and recording the resulting process variable (PV) response until it stabilizes at a new steady state. This curve is fit to a First-Order Plus Dead-Time (FOPDT) model: $G(s) = \frac{K_p e^{-Ls}}{\tau s + 1}$.

The 63.2% Rule is the standard mathematical method to identify the process Time Constant ($\tau$):

• Process Gain ($K_p$): $\Delta PV / \Delta CO$.
• Dead Time ($L$): The time delay between introducing the step change and the first observable change in PV.
• Time Constant ($\tau$): The time required for the PV to reach exactly 63.2% of its total steady-state change, measured from the end of the dead time.

Mathematical Basis: The response of a first-order system to step input $M$ is $y(t) = K_p M (1 - e^{-t/\tau})$. When $t = \tau$, $y(\tau) = K_p M (1 - e^{-1}) = K_p M (1 - 0.3678) = 0.632 K_p M$, which corresponds exactly to 63.2% of the final value.

Answer:

• Self-Regulating Process: Naturally settles to a new steady state when a step change is introduced in the controller output. For example, in a heat exchanger, opening the steam valve from 40% to 50% will raise the water temperature, which eventually stabilizes at a higher value (e.g. 75°C) because heat loss equals heat input.
  Transfer Function: $G(s) = \frac{K_p}{\tau s + 1}$.
• Integrating Process (Non-Self-Regulating): Does not stabilize. Instead, it continues to rise or fall at a constant rate for any steady input. The classic example is a liquid level tank. If the outlet valve is partially closed, water level continues to rise indefinitely until the tank overflows.
  Transfer Function: $G(s) = \frac{K_p}{s}$.

Tuning Rule: Integrating processes require highly stable tuning with conservative Proportional Gain ($K_c$) and relatively long Integral Times ($T_i$) to avoid ramping oscillations. Derivative action ($T_d$) is often extremely helpful for integrating loops to provide damping.

Answer: Flow and liquid level sensors (like differential pressure transmitters) are inherently prone to high-frequency measurement noise (e.g. from turbulences, pumps, or bubble formations). Because derivative action calculates the rate of change of this signal, any tiny high-frequency fluctuation in PV will produce a massive derivative response: $u_d(t) = K_c T_d \frac{d}{dt}[PV_{noise} \sin(\omega t)] = K_c T_d \omega PV_{noise} \cos(\omega t)$.

Since $\omega$ is very large for noise, the derivative output will swing wildly between 0% and 100%. This causes the control valve to chatter, leading to rapid wear of the actuator, seal failure, and high mechanical fatigue, without improving loop performance. Thus, flow and level loops are tuned almost exclusively as PI controllers.

When is Derivative Critical? It is essential in temperature control loops (large thermal inertia/lag $\tau$) and exothermic reactors. These processes have very slow dynamics, allowing the derivative action to predict thermal runaway and apply corrective action before the temperature overshoots safety limits.

Answer: Cascade Control involves using two PID controllers in series to control a single process variable. The Primary (Outer) Controller monitors the main process variable (e.g. tank temperature) and calculates the setpoint for the Secondary (Inner) Controller, which monitors a fast-responding variable (e.g. steam flow rate) and directly drives the control valve.

When to use it: Use cascade control when there are fast, recurring disturbances (like steam pipeline pressure fluctuations) that would affect the primary variable before the primary controller can detect and correct them. The secondary controller rejects these flow rate disturbances immediately, before they can alter the temperature.

Tuning Rule: Always tune the Secondary (Inner) loop first in manual or normal mode, with the primary controller in manual. Because the inner loop must be extremely fast to respond, tune it aggressively (typically PI or P only). Once the inner loop is stable, put it in Auto (Cascade mode) and tune the Primary (Outer) controller. As a rule of thumb, the inner loop's bandwidth must be at least 5 to 10 times faster than the outer loop.

Answer: Feedback control is reactive — it waits for a disturbance to affect the process, create an error ($SP - PV$), and then adjusts the controller output. Feedforward control is proactive — it measures the disturbance directly before it can affect the process, calculates the required corrective action, and adds it to the valve signal immediately. This prevents the error from occurring in the first place.

Mathematical Transfer Function:

To cancel out a disturbance transfer function $G_d(s)$ through the process control action $G_p(s)$, the ideal Feedforward controller $G_{ff}(s)$ is calculated as:

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

If $G_d(s) = \frac{K_d}{\tau_d s + 1}$ and $G_p(s) = \frac{K_p}{\tau_p s + 1}$, the feedforward block is a Lead-Lag compensator: $G_{ff}(s) = - \frac{K_d}{K_p} \left( \frac{\tau_p s + 1}{\tau_d s + 1} \right)$.

Answer: Gain Margin (GM) and Phase Margin (PM) are frequency-domain metrics that measure how close a closed-loop system is to becoming unstable. An unstable loop is one where the open-loop gain is $\ge 1$ (0 dB) at the exact frequency where the phase delay is $-180^\circ$ (which causes positive feedback).

• Gain Margin (GM): The factor by which loop gain can be increased before the system becomes unstable. Measured at the crossover frequency where phase lag is exactly $-180^\circ$.
  Industrial Target: $GM \ge 2.0$ (6 dB) is standard. For critical processes (e.g. exothermic reactors), $GM \ge 3.0$ is preferred.
• Phase Margin (PM): The additional phase lag required to make the system unstable. Measured at the gain crossover frequency where loop gain is exactly $1$ (0 dB).
  Industrial Target: $PM \ge 30^\circ$ to $45^\circ$. Lower phase margins produce excessive overshoot and oscillation.

Answer: When a process has a large dead time ($L > \tau$), standard PID controllers fail because the controller output adjustments do not affect the process variable until after the delay. Any aggressive correction winds up causing severe loop oscillations. The Smith Predictor is a model-based control scheme that bypasses this limitation.

It places a mathematical model of the process *without delay* ($G_p(s)$) and *with delay* ($G_p(s)e^{-Ls}$) inside the controller feedback path. The controller receives feedback from a simulated, delay-free process immediately, allowing the gain to be tuned tightly as if the dead time did not exist. The actual process delay feedback is compared with the model delay feedback to correct for modelling errors.

Limitations: It is highly sensitive to model mismatch. If the actual process gain $K_p$ or dead time $L$ changes slightly during operation (e.g. from valve scaling), the predictor cancellation will fail and the system will oscillate or become unstable.