A servo can be commanded to hold a position or follow a path, but how firmly it holds and how faithfully it follows depend on how its control loops are tuned. Tuning is the process of setting the loop gains, and raising those gains makes the axis stiffer and quicker, resisting disturbances and tracking commands more tightly. The catch is that gains cannot rise without limit: the load's inertia and the machine's mechanical resonances eventually push back and force a compromise. This guide explains how tuning raises stiffness and bandwidth, why inertia ratio and resonance cap the gains, and how notch filters and autotuning help push those limits.
Servo Loop Tuning in one line: Servo loop tuning is the setting of a servo's loop gains to make the axis respond quickly and accurately, and stiffness is the result: how firmly the axis resists being pushed off its commanded position or path. Raising the gains increases stiffness and bandwidth, but the load-to-motor inertia ratio and the machine's mechanical resonances limit how high the gains can go before the system becomes noisy or unstable, so tuning is a balance aided by notch filters and autotuning.
Stiffness in a servo means how hard the axis fights to stay where it is commanded when something tries to push it away. A stiff axis, disturbed by a cutting force, a load change, or a knock, produces a strong corrective torque immediately and barely deflects, while a soft axis gives way and takes time to recover. That stiffness comes directly from the loop gains: a higher gain means the loop reacts more forcefully to a given error, so the axis resists disturbance harder and returns to target faster. Tuning up the gains is, in effect, tuning up the stiffness.
Bandwidth is the companion idea in the time domain. It describes how fast the loop can respond to changes, whether following a rapidly moving command or rejecting a fast disturbance. Higher gains raise bandwidth, so a well-tuned, high-gain axis tracks quick moves closely, settles rapidly at the end of a move, and shrugs off disturbances before they grow. Low gains give a sluggish axis that lags its commands and lets disturbances linger. Stiffness and bandwidth rise together as gains increase, which is why tuning for performance is largely about pushing the gains as high as the machine will allow.
The gains being tuned live mostly in the nested velocity and position loops. The velocity loop's proportional and integral gains set most of the axis's stiffness against disturbances, and the position loop's gain and any feedforward set how tightly it tracks a commanded path. The tuning process generally works from the inside out, establishing a stiff, stable velocity loop first and then closing and tuning the position loop on top of it, because the outer loop can only be as good as the inner loop it depends on.
If gains could rise forever, every servo would be infinitely stiff, but two mechanical realities set a ceiling. The first is the inertia ratio, the ratio of the load's inertia to the motor's own. When the load's inertia greatly exceeds the motor's, the motor has a heavy, sluggish mass to control, and the same gain that is comfortable with a light load becomes twitchy and prone to oscillation with a heavy one. A high inertia ratio therefore forces lower gains and a softer, slower axis, which is why matching motor to load inertia is a fundamental part of sizing a servo system for good performance.
The second and often harder limit is mechanical resonance. No mechanism is perfectly rigid; couplings, shafts, belts, and structures flex, and every flexible mechanism has natural frequencies at which it wants to vibrate. As the loop gains rise and the bandwidth approaches one of these resonant frequencies, the loop begins to excite the resonance instead of controlling it, and the axis breaks into a high-pitched buzz or oscillation. This resonance is usually the first thing to appear as gains are pushed up, and it caps the achievable bandwidth well below what the electronics alone could deliver.
Because of these limits, tuning is a search for the highest gains that still leave a safe margin before instability, expressed as gain and phase margin. Push too far and the axis rings, buzzes, or oscillates; stay too conservative and it is soft and slow. The right tuning sits as high as the inertia and the mechanics allow while keeping enough margin that normal variations, a change in load, temperature, or wear, do not tip the axis into instability. This is why a servo cannot simply be set to maximum gain, and why understanding the mechanics is as important as adjusting the numbers.
When a specific mechanical resonance is what limits the gains, a notch filter is the classic remedy. A notch filter is placed in the control loop and tuned to remove the narrow band of frequencies at the resonance, so the loop no longer excites that mode. With the troublesome frequency filtered out, the gains can often be raised further before the next limit appears, buying more stiffness and bandwidth than the raw mechanics would otherwise allow. Identifying the resonant frequency, often from a frequency-response measurement of the axis, is the key to placing the notch correctly.
Because good tuning requires understanding the machine's dynamics, most modern drives offer autotuning to do much of the work. An autotuning routine typically excites the axis, measures how it responds, estimates the inertia and identifies resonances, and sets the gains and filters accordingly. Autotuning gives a solid starting point quickly and handles the common cases well, though demanding or unusual machines often still benefit from an engineer refining the result by hand, especially where the load inertia changes during operation or several resonances interact. Either way, the aim is the same: the stiffest, most responsive axis the mechanics will support with adequate stability margin.
The tuning itself is done in the drive, but whether an axis stays well-tuned over its life is something a plant can watch through cloud SCADA. Tuning is not entirely set and forget: as a machine wears, as loads change, and as mechanisms loosen, an axis that was crisp can drift toward sluggishness or start to buzz, and following error, settling behaviour, and torque demand shift accordingly. A platform such as Merobix can trend those performance indicators reported by drives across many machines and sites, so an axis whose tracking is degrading or that is starting to ring reveals itself in the data. For equipment spread across remote sites, catching that drift centrally lets a team schedule a re-tune or investigate a developing mechanical problem before the axis's declining stiffness turns into scrap or a breakdown.
Because the machine's mechanics fight back. Raising gains increases stiffness and bandwidth, but a high load-to-motor inertia ratio makes the axis twitchy, and every real mechanism has resonant frequencies that the loop begins to excite as gains rise, breaking into buzz or oscillation. Good tuning finds the highest gains that still leave a safe stability margin, so the axis stays stiff and responsive without ringing when the load, temperature, or wear varies.
The inertia ratio is the load inertia divided by the motor inertia, and it strongly influences how high the gains can go. A low ratio, where the motor's inertia is comparable to the load's, is easy to tune stiffly. A high ratio means the motor is controlling a much heavier mass, which makes the same gains oscillation-prone and forces lower gains and a softer axis. Matching motor to load inertia during system sizing is a key part of achieving good, stiff tuning.
A notch filter removes a narrow band of frequencies from the control loop, tuned to a specific mechanical resonance that is limiting the gains. By filtering out that frequency, the loop no longer excites the resonance into oscillation, so the gains can often be raised further for more stiffness and bandwidth. Placing the notch correctly requires knowing the resonant frequency, which is usually found from a frequency-response measurement of the axis.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.