What are the differences between PID and other control strategies?

PID and other control strategies

PID vs Other Control Strategies: Feedback Is Only One Way to Control a Process

PID is often the first controller considered in industrial automation because it is practical, well understood and available in almost every PLC, process controller and modern drive.

That does not mean every control problem should be reduced to three gains.

Some systems benefit from acting before an error develops. Others change so much during operation that one fixed set of gains is not enough. In multivariable plants, several process variables may have to be controlled together while respecting hard physical limits.

For those reasons, feedforward control, cascade control, gain scheduling, adaptive control, state feedback and model predictive control may be used instead of or, very often, together with PID feedback.

The useful question is not “Which controller is the most advanced?” but “What information is available, what is changing, and what problem must the controller solve?”

PID waits for measurable error

A conventional PID controller is fundamentally a feedback controller.

It compares the setpoint with the measured process variable and responds to the resulting error.

Suppose a pump is expected to maintain a discharge pressure of 6 bar. If demand increases and pressure falls to 5.5 bar, the PID controller sees the pressure error and increases pump speed.

This is extremely useful because the controller does not need to know exactly why the pressure changed. It only needs a reliable measurement of the result.

The familiar P, I and D actions then determine how strongly and how quickly the controller responds.

The weakness of feedback is also obvious: something must change first

Feedback normally responds after a disturbance begins to affect the controlled variable.

If a cold product suddenly enters a heated process, the temperature controller has to see some temperature deviation before feedback can react.

If a large consumer opens on a compressed-air or water network, pressure has to start changing before a pressure-only PID loop knows anything has happened.

That does not make feedback poor control. It simply identifies a situation where additional information may improve the response.

Feedforward acts on information before the controlled variable moves

Feedforward control uses information about a known command or measurable disturbance to calculate an additional control action.

Consider a heated production line where the incoming material flow rate is measured.

If material flow suddenly increases, the control system already knows that more heat will probably be required. A feedforward term can increase heater demand before the measured product temperature has significantly fallen.

The same idea can be used in pumping.

If process flow is measured independently, an increase in demand can contribute directly to the pump-speed command before a large pressure error develops.

Feedforward attempts to prevent predictable error. Feedback corrects the error that remains.

This is why the two strategies are frequently used together rather than chosen as competitors.

Why feedforward alone is rarely enough

A pure feedforward controller depends heavily on what it knows about the process.

It may compensate correctly for measured flow but know nothing about:

pipe restriction,

filter contamination,

motor deterioration,

unknown heat loss,

sensor bias,

mechanical wear.

A process model is also never perfectly accurate.

Feedback provides the correction mechanism for these modelling errors and unmeasured disturbances.

This gives one of the most useful industrial structures:

Feedforward command + PID feedback correction

The two functions have different jobs.

Servo systems are a good example of feedback plus feedforward

In a servo system, the controller already knows the requested trajectory, velocity and acceleration.

Velocity or acceleration feedforward can use this command information to produce part of the required motor effort before a following error becomes large.

The encoder-based feedback loops then correct the remaining error caused by load changes, friction, modelling errors and external disturbance.

For the feedback-loop side, see PID control with servo motors.

A well-designed feedforward term reduces the amount of work feedback has to do; it does not remove the need for feedback.

Two-degree-of-freedom PID is another variation

Not every advanced-looking controller is completely separate from PID.

A two-degree-of-freedom PID structure can treat setpoint changes differently from disturbance rejection by weighting the proportional and derivative response to the reference.

This can help reduce aggressive output movement after a setpoint step while preserving good disturbance rejection.

It is useful to distinguish this from disturbance feedforward. Both can improve transient behaviour, but they use different information and solve different problems.

Cascade control: one controller gives the setpoint to another

Some processes contain a fast inner variable that can be controlled before a slower outer variable is affected.

This is where cascade control can be useful.

For example, a slow product-temperature controller may generate the setpoint for a faster jacket-temperature controller.

The outer controller cares about product temperature.

The inner controller deals with rapid changes in heating conditions.

For cascade control to work well, the inner loop should normally respond significantly faster than the outer loop.

Connecting two PID blocks in series does not automatically create a good cascade controller.

The process dynamics and tuning of both loops must support the architecture.

The nested position, speed and torque/current loops found in motion systems are another familiar example of controllers operating at different dynamic levels.

Gain scheduling: different operating regions, different gains

A machine may behave very differently at low and high speed.

A process may also change significantly with temperature, pressure, product type or load.

One practical solution is to prepare more than one set of controller gains.

For example:

Low-speed operation: tuning set A

Normal operation: tuning set B

High-load operation: tuning set C

The controller selects or interpolates between gain sets according to the current operating region.

This is known as gain scheduling.

It should not automatically be described as adaptive control. The gains may have been completely designed in advance, with no online learning or parameter estimation taking place.

Adaptive control changes its own control behaviour online

Adaptive control becomes relevant when important plant characteristics are uncertain or change during operation.

A robot is an easy example.

The dynamics of an empty robot arm are not identical to those of the same arm carrying a heavy payload. Likewise, a process may change as product viscosity, temperature or production rate changes.

An adaptive controller can update controller parameters or estimates of plant behaviour using information collected during operation.

This is broader than “automatic PID tuning”.

Some adaptive strategies may modify feedback and feedforward gains. Others estimate plant parameters or disturbance models and calculate control action from the updated model.

In model-reference adaptive control, for example, controller or plant estimates can be updated from real-time tracking error so that the actual system follows a desired reference behaviour.

For a practical motion example, compare this with PID control in robotics.

Adaptive control is not the same as auto-tune

This distinction matters.

A PID auto-tune procedure may test a process during commissioning, calculate suitable gains and then leave those gains unchanged during normal production.

An adaptive controller, by contrast, can continue updating selected parameters while the plant operates.

Therefore:

Auto-tune: tuning procedure

Gain scheduling: predefined parameter changes

Adaptive control: online adjustment based on changing plant behaviour or performance

They may overlap in particular products, but they should not be treated as identical concepts.

State feedback uses more information about the system

PID usually focuses on the relationship between a reference and a measured controlled variable.

State-feedback control can instead use several internal state variables of a dynamic system at the same time.

A mechanical system, for example, may be represented through quantities such as position and velocity rather than only a single scalar control error.

This gives the designer direct control over the dynamics represented by the model.

The trade-off is that a useful mathematical model and the required state measurements or state estimates are needed.

PID remains attractive precisely because many practical processes can be controlled successfully without building a detailed state-space model.

Model Predictive Control looks ahead

Model Predictive Control, usually abbreviated as MPC, uses a process model to predict how the plant will behave over a future prediction horizon.

At each control step it calculates a sequence of possible future control moves and solves an optimization problem. Only the first selected control move is applied, and the calculation is repeated at the next step using updated measurements.

This approach is particularly useful when several manipulated and controlled variables interact.

MPC can also explicitly incorporate restrictions such as:

maximum valve opening,

minimum pump flow,

heater power limit,

rate-of-change limit,

maximum permitted process temperature.

Handling such input and output constraints is one of MPC's major advantages. Current MPC implementations also naturally support multivariable systems and can use known future reference or disturbance information.

MPC is powerful, but power has a cost

A standard PID calculation is inexpensive and easy to understand.

An MPC controller needs a process model, prediction calculations and an optimization solution.

That means more engineering effort, more computing resources and usually more specialised maintenance knowledge.

For a complicated refinery or strongly interacting process, that effort may be justified.

For one temperature loop controlling one heater, it may be unnecessary.

Using a more complicated controller than the process needs is not good engineering.

ON/OFF control is still the right answer for some machines

There are also systems where PID is already more complicated than necessary.

A basic thermostat can turn a heater ON below one temperature and OFF above another, with hysteresis preventing rapid switching.

This approach can be perfectly suitable when the acceptable control band is relatively wide and the process is slow.

PID becomes attractive when tighter regulation, smoother control output or reduced cycling is important.

The simplest method that satisfies the real process requirement is often the most maintainable one.

Example: pressure control with measurable demand

Consider a plant where a pressure transmitter measures supply pressure and a VFD controls pump speed.

A normal feedback PID loop increases motor speed after pressure begins to fall.

If process demand is also measured with a flowmeter, that flow value can be used as feedforward information.

A sudden demand increase can then produce an immediate additional speed command.

The pressure PID remains active and corrects whatever pressure error remains.

The process loop itself is discussed in PID pressure control, while the drive implementation is covered in PID control with a VFD.

Example: flow disturbance in a temperature process

Suppose a heat exchanger must maintain outlet temperature.

With feedback alone, the temperature controller reacts after outlet temperature deviates.

If inlet flow is measured, a feedforward term can estimate the extra heating or cooling required when flow changes.

The temperature feedback loop then corrects model error, fouling, inlet-temperature changes and other effects not perfectly represented by feedforward.

This is a much more realistic industrial use of feedforward than attempting to replace the temperature feedback loop entirely.

Example: changing payload in robotics

A robot tuned for an empty gripper can behave differently after picking up a heavy component.

If the controller has an accurate payload model, gravity and dynamic compensation can predict some of the additional required torque.

If the system also includes adaptive functions, selected parameters may be updated when plant behaviour changes.

The lower-level joint feedback still remains necessary to correct disturbances and modelling error.

This is why model-based compensation, adaptive control and PID feedback can exist in the same robot.

When does ordinary PID start to struggle?

A well-designed PID controller can solve a remarkable number of industrial problems, but PID has limitations.

Alternative or additional control methods become worth considering when the plant has:

very long dead time,

large changes in dynamics across the operating range,

strongly interacting variables,

important measurable disturbances,

tight actuator and process constraints,

large nonlinearities,

changing or uncertain plant parameters.

This does not automatically mean PID must be removed. It may mean PID needs help from another layer.

How should a control strategy be chosen?

Start with the process, not with the algorithm name.

Can the controlled variable be measured reliably?

Can important disturbances be measured before they affect the output?

Does the plant change significantly with operating point or load?

Are several manipulated variables and outputs strongly coupled?

Are actuator and process constraints central to the problem?

How much dead time is present?

How fast must the controller execute?

Who will maintain and troubleshoot the controller five years from now?

Those questions usually narrow the choice considerably.

A practical comparison

ON/OFF control: appropriate when two-state operation and a reasonable dead band are sufficient.

PI or PID: usually the first serious option for a conventional single-loop process with reliable feedback.

Feedforward + PID: useful when an important disturbance can be measured before the controlled variable reacts.

Cascade control: useful when a faster secondary process variable can be controlled inside a slower primary loop.

Gain scheduling: practical when good controller settings are known for several distinct operating regions.

Adaptive control: useful when important plant behaviour changes or remains uncertain during operation.

State feedback: attractive when a useful dynamic model and state measurements or estimates are available.

MPC: especially valuable when several variables interact and constraints must be considered explicitly.

Can a PLC use these strategies?

Yes, depending on PLC performance, programming environment and manufacturer libraries.

A PLC can implement more than a basic PID loop. PLC control applications can include cascade loops, feedforward terms, operating-region logic and custom control calculations.

More computationally demanding model-based strategies may instead run in a high-performance controller, industrial PC or dedicated process-control platform.

Regardless of the algorithm, sampling time, I/O update rates and deterministic execution still matter. This is covered further in real-time PID control.

Advanced control does not replace safety engineering

A controller may use PID, feedforward, MPC or adaptive algorithms and still remain a normal process-control system.

None of these methods automatically provides the independent protective functions required for dangerous pressure, temperature, mechanical movement or personnel hazards.

Better process control and machine safety are separate engineering responsibilities.

Related PID and Control Topics

PID fundamentals: PID Control FAQ - What Is PID? - P, I and D Components - PID Limitations

Dynamics and tuning: PID for Different Process Types - Time Delay and Process Dynamics - PID Gain and Tuning

Control platforms: PID with PLCs - PID with Servo Motors - PID in Robotics - Real-Time PID Control

Process applications: Pressure Control - Flow Control - Level Control - PID with VFDs

The best control strategy is the one that meets the real process requirement with acceptable performance, reliability and maintainability. Greater algorithmic complexity should have a clear engineering reason.