Este guia está disponível atualmente em inglês.

Technology / Technical guide

Control Saturation, Integral Windup & Anti-Windup

A controller can request more than an actuator or conversion stage can deliver. If its integral state keeps accumulating the unmet error, the stored request can remain excessive after conditions change. This is a different phenomenon from the physical limit itself.

Clipping an output does not necessarily limit the controller’s memory.

Separate requested action, constrained action and accumulated state to understand windup and recovery.

01 / Physical limit

A requested quantity can exceed its available range.

Saturation means that an output stops following the unconstrained request once a defined limit is reached.

Saturation means that an output stops following the unconstrained request once a defined limit is reached. A current, torque or other output can have a bounded range. A rate limit instead restricts how quickly it changes; these constraints need not be the same.

Hypothetical amplitude example: a mathematical request of 12 units passing through a limiter capped at 8 units produces 8 units. The four-unit difference concerns the requested and limited quantities, not proof of a failed actuator.

02 / Integral memory

Unmet error can keep accumulating.

If a controller continues integrating a same-sign error while the applied output is already limited, its integral contribution can grow beyond what the physical path supports.

If a controller continues integrating a same-sign error while the applied output is already limited, its integral contribution can grow beyond what the physical path supports. The observed error may persist because more requested action cannot produce more applied action.

MathWorks illustrates this windup mechanism with saturated control models. A limiter can protect the output boundary in that model while leaving the controller’s internal state excessive. Output clipping and state management therefore answer different questions.

03 / Recovery effect

A changed error may first unwind stored history.

After the reference or physical condition changes, the controller may need opposite-sign accumulation before its request re-enters the achievable range.

After the reference or physical condition changes, the controller may need opposite-sign accumulation before its request re-enters the achievable range. During that period, the limited output can continue to differ from the response implied by the new current error.

Recovery delay is not uniquely evidence of windup: plant inertia, filtering and other dynamics can also delay response. The mechanism is understood by separating error, integral state, unconstrained request and applied quantity, rather than judging one late-looking trace alone.

04 / Anti-windup role

State management responds to constrained action.

Anti-windup methods aim to keep integral behaviour consistent with the output constraint.

Anti-windup methods aim to keep integral behaviour consistent with the output constraint. Conditional integration changes when accumulation proceeds; back-calculation uses information about the discrepancy between unconstrained and constrained action to influence the internal state.

These are alternative conceptual structures, not universal remedies with identical results. Their appropriate representation depends on where the limit occurs and which information is available. Managing integral state cannot create extra torque, current or available source power.

05 / Several boundaries

A downstream constraint may differ from an internal limiter.

An internal controller limit can differ from a limit further along the actuator or conversion path.

An internal controller limit can differ from a limit further along the actuator or conversion path. If the later applied quantity is unknown, a controller may not have the information assumed by a particular anti-windup model. Cascaded loops can also encounter different limits.

The PI/PID guide defines the stored integral contribution, while the feedback guide distinguishes a request from physical response. Keeping these boundaries clear explains why one clipped signal is not the whole loop.

06 / Product scope

Anti-windup is a defined function, not an inferred feature.

The presence of a controller, actuator limit or integral concept does not establish an anti-windup implementation in MARSTON equipment.

The presence of a controller, actuator limit or integral concept does not establish an anti-windup implementation in MARSTON equipment. Nor does a model’s faster recovery establish actual stable or safe turbine behaviour under every constrained condition.

No saturation value, reset command, gain, clamping rule or controller modification is supplied. The invented limiter is a mathematical example only, and actual fault or protection behaviour remains a separate equipment-specific subject.

Continue the review with PI & PID Control: Proportional, Integral & Derivative Roles.

Continue the review with Yaw Feedback: Setpoint, Measurement & Closed-Loop Error.

Continue the review with Power Device Safe Operating Area: Conditions and Duration.

General background: MathWorks: Integral windup, conditional integration and back-calculation concepts. These references explain industry concepts; they are separate from the applicable MARSTON product and project documents.

Key concepts and conditions.

Read each quantity with its reference, units and stated conditions. A general relation does not establish a model-specific value.

Concept / quantityMeaning / relationConditions / limits
Unconstrained requestRequested u with its unitsCan differ from physical output.
SaturationBounded output rangeAmplitude and rate constraints are distinct.
Integral stateAccumulated error and existing stateOutput clipping need not stop accumulation.
RecoveryRe-entry into achievable request rangeSeveral dynamic causes can affect timing.
Anti-windupState response to actual limitingCannot add physical source or actuator capacity.

Engineering questions.

Is output clipping the same as anti-windup?

No. Clipping can leave the integral state accumulating.

Is every slow recovery caused by windup?

No. Other dynamics and measurement effects may contribute.

Does anti-windup increase available torque?

No. It manages state around a constraint rather than removing physical limits.

Are amplitude and rate limits identical?

No. They restrict magnitude and rate of change respectively.

Does a limiter prove the real controller uses back-calculation?

No. That requires actual implementation evidence.

Are the 12 and 8 units product limits?

No. They belong to an invented mathematical limiter.

Technical enquiry

Discuss the technical question.

A controller can request more than an actuator or conversion stage can deliver. Share the model and applicable operating or design information for an engineering discussion.