Cascade and Feedforward Control

Core Idea

Cascade control uses two nested feedback loops to improve disturbance rejection and response speed: an inner (secondary) loop with a fast sensor controls an intermediate variable, while an outer (primary) loop sets the inner loop's setpoint based on the primary controlled variable. The inner loop rejects disturbances entering the secondary process before they propagate to the primary output, and it linearizes the inner process dynamics as seen by the outer controller. For cascade control to be effective, the inner loop must be significantly faster than the outer loop (typically 3-5 times faster) so the outer controller can treat the inner loop as approximately unity gain. Feedforward control takes a fundamentally different approach: it measures a disturbance directly and applies a corrective control action before the disturbance affects the output, using a feedforward transfer function G_ff = −G_d/G_p (where G_d is the disturbance-to-output path and G_p is the control-to-output path). Perfect feedforward cancellation requires exact knowledge of G_d and G_p, which is never available in practice, so feedforward is almost always combined with feedback to handle modeling errors and unmeasured disturbances. Combined cascade-feedforward architectures are common in process control, where the feedforward signal adjusts the inner loop setpoint while the outer loop corrects for residual errors.

How It's Best Learned

Simulate a heat exchanger or tank-level process with a measurable disturbance (e.g., inlet temperature or flow). First implement single-loop PID control and observe the disturbance response. Then add an inner cascade loop around the fast actuator dynamics and compare. Finally, add feedforward from the measured disturbance and observe the incremental improvement. This layered comparison makes the value of each architecture tangible.

Common Misconceptions

• Cascade control does not require two different physical variables — it requires two measurements at different time scales, but in some applications the inner and outer measurements are the same variable at different points in the process.
• Feedforward control cannot work alone in practice because it provides zero correction for unmeasured disturbances, model uncertainty, and sensor drift — it is a complement to feedback, not a replacement.
• The inner loop in a cascade must be tuned first with the outer loop open; tuning both loops simultaneously couples their dynamics and typically leads to oscillatory or unstable behavior.

Explainer

Standard PID feedback control responds to error — it waits for a disturbance to push the controlled variable away from setpoint, then corrects. The fundamental limitation is that correction always lags behind the disturbance: by the time the controller acts, the output has already moved. Cascade control and feedforward control are two structurally different strategies for dealing with this lag, and understanding each requires seeing clearly what the single-loop feedback structure cannot do.

Cascade control attacks a specific vulnerability: disturbances that enter the process partway through the chain between actuator and output. Consider a heat exchanger where you control outlet temperature by adjusting a steam valve. A disturbance in steam supply pressure changes the flow through the valve before it affects the outlet temperature. Single-loop control on outlet temperature waits for the temperature to drift before correcting. A cascade architecture adds an inner loop that measures steam flow directly and keeps it at the setpoint commanded by the outer loop on temperature. The inner loop rejects the pressure disturbance at the flow level — before it ever reaches the thermal dynamics. The outer loop then sees what appears to be a well-behaved, fast actuator that precisely delivers whatever flow rate it commands. This is the linearization benefit: the inner loop absorbs the nonlinear, uncertain actuator dynamics, making the outer loop's control problem simpler. The critical constraint is timing — the inner loop must close roughly 3–5 times faster than the outer loop, or the outer controller cannot treat the inner loop as instantaneous and the separation breaks down.

Feedforward control takes an entirely different approach: instead of waiting for error, it measures a known disturbance *before* it affects the output and pre-emptively applies a corrective signal. The ideal feedforward transfer function is G_ff = −G_d / G_p — the negative ratio of the disturbance-to-output path over the control-to-output path. If this ratio is perfectly modeled, the feedforward signal exactly cancels the disturbance effect. In practice, perfect cancellation is impossible (model uncertainty, unmeasured disturbances, time delays), so feedforward is always paired with a feedback loop that corrects the residual error. The combination is powerful: feedforward handles the known, predictable part of the disturbance, while feedback handles the random and unmodeled residual. The result is significantly faster disturbance rejection than feedback alone, without requiring a faster feedback loop or higher gains (which would compromise stability margins).

The architectures are complementary and frequently combined. A full process-control loop might use cascade control (inner flow loop, outer composition loop) *plus* feedforward from a measured feed composition disturbance. Each layer adds a targeted improvement: cascade improves rejection of disturbances that enter the inner process; feedforward eliminates the lag in responding to measurable disturbances. Both rely on your PID foundations — the individual controllers within the architecture are still PID regulators — but they restructure information flow to act earlier and more specifically than a single-loop design can manage.