Tuning and Troubleshooting Control Loops

A tuning procedure

1. Check the basics first. Is the sensor right and calibrated, filtered correctly and wired to the right input? Does the actuator move the right way, with no stiction and enough range? Is the sample time short enough (Digital Control and Sampling)? Controllers cannot cure a broken instrument.
2. Put the loop in manual at a steady operating point and let the process settle.
3. Make a step test of 5 to 10 percent in the output (large enough to see above the noise, small enough to stay in the linear range), and record the response until it settles.
4. Fit K, τ and θ from the response (below), then calculate the settings with a lambda, SIMC or reaction-curve rule (PID Control and Tuning).
5. Set anti-windup and the derivative filter, then switch to automatic and test with a small setpoint step and a disturbance.
6. Refine if needed: slower and smoother (a larger λ) if the loop rings or the process varies; faster only if the margin allows (Frequency Response and Bode Plots).
7. Record the settings, the date and the operating conditions. Retune when the process changes.

The step test

For a first-order-plus-dead-time process the response to a step Δu in the output rises after the dead time θ towards a final change Δy. Three numbers describe it: the gain K = Δy ÷ Δu, the dead time θ (when the response starts) and the time constant τ (how long to reach 63 percent of the way after the dead time). A robust way to read them from noisy data is the two-point (Smith) method: find the times t₂₈ and t₆₃ at which the response reaches 28.3 and 63.2 percent of the final change; then τ = 1.5 (t₆₃ − t₂₈) and θ = t₆₃ − τ.

Worked example: from data to settings

Diagnosing a bad loop

Look at the process variable, the setpoint and the controller output together in a trend, over a few minutes to hours, and ask three questions:

1. Is it oscillating? Measure the period. If it is about the ultimate period (for the oven P_u = 57 s, close to four dead times, 60 s), the gain is too high. If the period is several time constants long, the integral action is too strong (the loop is chasing the setpoint too slowly and overshooting).
2. What shape is the oscillation? A smooth sine in the PV and the output: a tuning problem. A triangular wave in the PV with a square-wave output: stiction or backlash. A sharp jump when the setpoint is moved: derivative kick.
3. Is it slow? A long creep to the setpoint with no overshoot: the gain or the integral is too small, or the output is saturated.

Symptoms and causes

Symptom	Likely cause	What to do
Steady oscillation with a sine-like shape, period near 60 s	gain too high; delay increased	reduce Kp (halving it restores a comfortable margin); check the sensor and filter delays
Slow oscillation, period many times the process lag	integral too strong (Ti too short)	lengthen Ti; for a lag-dominant process Ti ≈ τ is enough
Triangular PV, square output, fixed amplitude	valve stiction or backlash	fix the valve or positioner; do not detune the loop (Practical Control: Windup, Feedforward and Cascade)
Large overshoot after a long saturation	windup	enable anti-windup; limit the setpoint step or ramp it
Output noisy, actuator chatters	derivative on a noisy signal; high gain; filter too weak	filter the derivative, or use PI; shield and filter the measurement
Spike when the setpoint changes	derivative kick, proportional kick	derivative on the measurement; setpoint weighting
Works well at one load, oscillates at another	nonlinear valve or process gain	tune at the worst case; gain scheduling; change the valve characteristic
Sluggish response but no overshoot	gain too low, or output saturated, or manual	raise the gain; check the output limits and the mode
Offset that never disappears	no integral term; integral clamped; output limit reached	enable integral; check saturation and any “high select” override
Loop got worse after a maintenance job	sensor replaced (a different lag), actuator replaced, process change	rerun the step test and retune

Stiction and limit cycles

A control valve sticks when the friction of the packing is larger than the force the actuator can apply for a small change of signal. The controller then increases its output (through the integral term) until the valve breaks away and jumps, overshoots, and the process swings the other way: the result is a limit cycle that no tuning can remove (making the loop slower only lengthens the period). The amplitude is set by the stick band: a stick-slip of 1 percent of valve travel moves the oven by about K × 1 = 2 °C. A positioner on the valve, a smaller packing friction, or a servo-driven actuator cures it. The signature is a flat-topped, square-like controller output with a sawtooth or triangular process variable: plot OP against PV and look for the parallelogram.

Loop performance and maintenance

Industrial audits regularly find that a large share of control loops (often a third or more) are in manual, badly tuned or limited by a faulty valve or sensor. Simple metrics catch them: the time in automatic, the IAE or the standard deviation of the error against its history, the valve travel (a noisy loop moves a valve thousands of times a day: with the example PID at K_p = 4.8, N = 10 and 0.1 °C of noise the heater would be jittering by about 5 percent), the oscillation index and the output saturation time. Keep the settings in a register, retune after any change to the process or the instruments, and treat a loop that has been in manual for months as a fault to fix, not a feature.

Related

• Previous: Practical Control: Windup, Feedforward and Cascade. Back to the overview: Control Systems.
• PID Tuner for the settings from K, τ and θ.
• Sensors and Field Devices and System Integration for instruments and commissioning.

FAQ

How big should the step test be?

Large enough that the response is clearly above the noise (at least 5 to 10 times the noise amplitude) and small enough to stay linear: 5 to 10 percent of the output range is typical.

Should I retune after every change?

After any change that alters the process gain, lag or delay: a new sensor, valve, pump, product, or load range. Keep a record so a change is quick to compare.

Can auto-tune make things worse?

If the process is not at steady state, is disturbed during the test, or has a stiction problem, the numbers it finds are wrong. Check the settings against the physics before you accept them.

Why does the loop oscillate only in the morning?

Process conditions differ: a cold start, a different load, a different flow. A nonlinear process needs tuning for the worst case or a gain schedule.

References

• Åström, K. J., Hägglund, T., Advanced PID Control, ISA; Smith, C. A., Corripio, A. B., Principles and Practice of Automatic Process Control, Wiley (process reaction curve).
• Skogestad, S., Simple Analytic Rules for Model Reduction and PID Controller Tuning, Journal of Process Control 13 (2003).
• Choudhury, M. A. A. S., Shah, S. L., Thornhill, N. F., Diagnosis of Process Nonlinearities and Valve Stiction, Springer; IEC 61131-3 PID blocks and ISA-5.1 (instrumentation symbols).

Last reviewed: October 2026. Typical audit figures; the numbers in the example come from the oven model.