Derivative Action in PID Control Without Noise Problems

Derivative action is often the least understood part of a PID controller. Proportional action responds to the present error, while integral action accumulates past error. Derivative action responds to how quickly a selected signal is changing. Used carefully, it adds damping and helps a loop resist rapid motion toward overshoot. Used carelessly, it turns sensor noise and setpoint steps into sharp output movement.

The practical question is not whether derivative is good or bad. It is whether the process, measurement, controller implementation, and tuning objective give derivative useful information. A motion axis with measurable inertia may benefit from damping. A noisy flow signal may become worse when high-frequency variation is differentiated. Good tuning begins by identifying that difference.

What the Derivative Term Measures

For a controller error defined as setpoint minus process variable, derivative action is proportional to the rate of change of that error. If the error is changing rapidly, the derivative contribution is larger. If the error is steady, the derivative contribution approaches zero even when the error itself is not zero.

This behavior is why derivative can anticipate overshoot. Suppose a position loop is approaching its target quickly. The remaining error may be small, but its rate of change shows that the mechanism still has momentum. Derivative action can reduce the command before the axis crosses the target. It does not predict the future in a literal sense; it reacts to the current slope.

The derivative term responds to slope, not to the magnitude of a steady error.

Derivative on Error Versus Measurement

Controller implementations do not all place derivative in the same path. Derivative on error responds to both process-variable movement and setpoint movement. A sudden setpoint step can therefore produce a large temporary output change known as derivative kick. Derivative on measurement uses the negative rate of change of the process variable, avoiding a direct kick from a setpoint step.

Engineers must check the controller manual rather than assume one structure. The setting may be described as derivative on error, derivative on process variable, two-degree-of-freedom control, or setpoint weighting. The distinction matters during commissioning because the same gain can behave very differently after an operator changes a setpoint.

Why Measurement Noise Becomes Output Noise

A derivative emphasizes fast changes. Real sensors contain electrical noise, quantization, vibration, turbulence, and sampling variation. Even a small high-frequency component can create a noticeable derivative output. The result may be a chattering valve, a drive torque command that moves continuously, or an actuator that wears without improving process control.

Practical controllers therefore use a filtered derivative rather than an ideal mathematical differentiator. MathWorks' PID guidance explains why ideal derivative action amplifies high-frequency noise and why filtering is used in real systems. The filter reduces sensitivity above a selected frequency, but it also reduces the phase benefit that derivative was intended to provide. Filtering is part of the tuning tradeoff, not a cure that permits unlimited gain.

Sampling and Signal Quality Set the Limit

Derivative quality depends on the interval between samples. If the controller samples too slowly, it can miss meaningful dynamics. If it samples quickly but the signal resolution is poor, small count changes can look like steep slopes. Irregular task timing adds another error because the controller is dividing signal change by a changing time interval.

Before adding D gain, trend the raw process variable at the actual control-task rate. Confirm that scaling, timestamps, and engineering units are correct. Inspect the signal with the process steady and during a controlled change. If the trend is dominated by noise or aliasing, improve the measurement, sampling, shielding, grounding, or filtering first.

A Safe Tuning Sequence

Begin with a known-safe operating region and documented output limits. Disable or minimize derivative. Tune proportional and integral action to obtain a stable response that reveals the process dynamics. Then introduce a small derivative value while watching overshoot, settling time, output activity, and measurement noise.

Make one change at a time. Test both setpoint response and disturbance rejection because a tuning change may improve one while degrading the other. For motion systems, test across load, speed, and direction. For thermal or flow loops, test at more than one operating point because process gain and time constants may change with production rate.

Use controller features such as output clamps, slew limits, bumpless transfer, and mode-change tracking where available. The PLC and PAC systems catalog shows the range of controllers used for discrete, motion, and process applications, but the exact PID structure must always be confirmed in the model-specific documentation.

Symptoms of too much derivative

Excess D gain often appears as high-frequency output movement, sensitivity to sensor spikes, audible actuator activity, or alternating corrections near the setpoint. A loop may look well damped in a smooth simulation yet behave poorly on the machine because the real measurement contains noise and mechanical resonance.

Do not treat every oscillation as a request for more damping. Check mechanical backlash, loose couplings, valve stiction, poor sensor mounting, network delay, and task jitter. Derivative can mask or magnify these issues, but it cannot repair them.

Risk Boundaries During Commissioning

Derivative changes can create fast output transitions. Establish travel limits, torque or pressure limits, safe test amplitudes, and an accessible stop before experimenting. Trend the command, process variable, error, derivative contribution, and any saturation flags. If the controller does not expose the individual PID terms, use a test environment or reduced-energy condition to understand the response.

For loops that drive heavy mechanics, start with unloaded or reduced-speed tests when the machine design permits. For valves and heaters, consider equipment wear and thermal lag. Never tune around an active safety function or use the basic PID loop as a substitute for independent protective logic.

When PI Is the Better Choice

Many industrial loops perform well with proportional and integral action alone. Slow thermal systems, noisy flow measurements, and processes with significant dead time may gain little from derivative. Removing D reduces complexity and makes future maintenance easier. The right design is the simplest controller that meets the process objective across expected operating conditions.

Derivative action earns its place when it produces measurable damping without excessive output activity. That decision should be based on trends, repeatable tests, and knowledge of the controller structure. The engineering value of D is not a dramatic response on one demonstration. It is a stable improvement that remains understandable to the next person who services the loop.