After every sharp corner, embossed letter or hole edge, the wall shows a repeating ripple that fades out over the next centimetre or two. Run a fingernail across it and you can feel the waves. The part is dimensionally fine; the surface is an echo.
The machine is a tuning fork
The toolhead has mass and the frame that carries it has stiffness. Any structure with both has a natural frequency, and any sharp change in motion excites it — exactly like flicking the end of a ruler held over a desk edge.
When the toolhead stops or reverses abruptly, the frame flexes, springs back, and oscillates for a while at that frequency. The nozzle is attached to the frame, so it traces the oscillation into the surface it is printing. The ripples fade as the vibration damps, which is why the marking always appears after a feature and never before it.
The energy that starts the oscillation comes from the change in velocity, not the velocity itself. That is the practical headline: acceleration and jerk settings control this, and print speed barely does. It is why people who halve their speed and see no improvement conclude, wrongly, that their frame is beyond help.
Measure your machine's frequency
The failed print contains the number, and getting it takes calipers.
Measure the distance between two adjacent ripple crests, and note the speed the perimeter was printed at. The frequency is that speed divided by the spacing. Ripples 2 mm apart on a wall printed at 100 mm/s mean the frame is ringing at fifty cycles per second.
That number is worth having. It tells you whether the two axes differ — they usually do, since a bed-slinger's Y axis carries the whole plate and its mass while X carries only the toolhead. Firmware with resonance compensation asks for exactly this figure, and an accelerometer-based calibration measures it directly. If your firmware supports that, use it: the manual measurement is for machines that do not.
What to do, cheapest first
- Reduce acceleration, not speed, and reprint the same test object. A large reduction here costs surprisingly little print time on typical geometry, because most paths never reach top speed anyway.
- Enable input shaping or resonance compensation if the firmware has it, using a measured frequency rather than a guessed one. This is the change that lets a machine keep its speed.
- Tension both belts so the two sides of an axis sound alike when plucked. A slack belt lowers the effective stiffness of the whole system and drops the frequency into a range that is easier to excite.
- Put the machine on something heavy and rigid. A wire shelving unit is a spring in its own right, and a printer on one is a mass-spring system stacked on another. A paving slab on a solid bench, or a concrete tile under the machine, is a genuine improvement for the price of a coffee.
- Tighten the frame. On bed-slingers, the joint between the uprights and the base is the usual culprit; on machines that have been moved, everything is.
Reduce the mass if you can
Anything bolted to the moving assembly makes this worse, because frequency falls as mass rises and a lower frequency is both easier to excite and slower to damp. A heavy direct-drive extruder, a bulky fan duct, a camera mounted on the gantry — all of them show up in the surface. There is a limit to what can be done here on a fixed machine, but it explains why two printers with identical firmware settings ring differently.
Other repeating patterns, and why they are not this
Ripples that repeat evenly up the whole part with no relation to corners are not ringing; that is Z banding, and it has a mechanical period rather than a decaying one. A fine, uniform shimmer with no measurable spacing is salmon skin, which comes from the drivers rather than the structure. Ringing always has a feature it can be traced back to.
Materials make almost no difference
This is one of the few faults on this site with no material component worth mentioning. Every filament records the same oscillation. Glossy and silk grades show it more clearly because they reflect light along the ridges, and matte grades hide it — which changes how visible the defect is without changing anything about its cause.