How 3D-Printed Infill Patterns Affect Vibration

Infill is the internal structure of a 3D-printed part. Engineers usually choose it for strength and weight, but it also changes how a part vibrates, which matters for lightweight parts in machines, robots, and aircraft that see dynamic loads.

Earlier studies looked at individual parameters, but few compared several common infill geometries under the same test conditions. I compared rectilinear, gyroid, honeycomb, and grid infill, keeping the outside dimensions and the 20% infill density identical so only the internal geometry changed.

Every specimen is a 215 × 33 × 20 mm PLA cantilever printed at 20% infill, in one of four patterns: rectilinear, gyroid, honeycomb, or grid. The printed masses came out between 55.31 g and 55.67 g, so the samples were effectively identical apart from their internal geometry.

I picked the length with Euler–Bernoulli beam theory. Using solid-section PLA properties it predicts a first natural frequency near 117 Hz — and that number is deliberately an upper bound, because 20% infill removes most of the material in the middle of the beam and drops the effective stiffness well below a solid section. What mattered was whether the patterns would separate inside the accelerometer's usable bandwidth, and they did.

A RATTMMOTOR CNC spindle (12–100 VDC, up to 16,000 RPM) bolted to a mounting plate excites the beam. The clamp holds the specimen with four M4 bolts and rubber washers for an even, distributed grip rather than a single screw, and the base of the assembly sits on a 1.5 mm vibration-damping mat so the table doesn't feed noise back into the beam.

An ADXL345 at the tip, set to ±16 g, streams over I²C to an Arduino Uno at a 570 Hz output data rate; every sample is converted to g and logged with a timestamp. I ran two test types on every pattern: steady-state forced vibration, and a natural-response run where the spindle is brought up to speed, cut at t = 0, and the 3.7-second coast-down is recorded.

The AC-DC converter powering the spindle radiated enough interference to corrupt readings on the Arduino board, so I installed a copper plate between the converter and the Arduino as a shield. I built the spindle's power and control electronics into their own enclosure.

I wrote a Python pipeline with NumPy, SciPy, pandas, and Matplotlib that processes every trial automatically. It runs a Fast Fourier Transform to find each beam's dominant resonant frequency and computes RMS acceleration as a single measure of how much vibration the beam transmits.

For damping, it detects the peaks in each free-decay signal, fits an exponential decay envelope, and converts the decay constant and natural frequency into a damping ratio (ζ). All plots and CSV summaries are generated directly from the raw data.

Infill geometry made a big difference. Grid was by far the stiffest, resonating at 83.1 Hz, more than double the other patterns, and it transmitted the least vibration, at about 1.2 g RMS versus 3.0–3.8 g for the others. It also showed the highest damping ratio.

Gyroid had moderate damping and mid-range transmission, while rectilinear and honeycomb had low damping and kept vibrating longer after excitation. In practice, this means a designer can tune how a printed part responds to vibration just by changing its infill, without changing its outside shape or weight.

None of the beams came near the 117 Hz solid-beam prediction, which is the expected result: with 20% infill the effective flexural stiffness is far below a solid section, so the real first modes landed in the 35–83 Hz range instead.

I co-authored the 20-page manuscript with my mentor, Tracy Lu, a Caltech mechanical engineering PhD candidate, through the Polygence research program. We submitted it to the National Journal of High School Science, and it came back with revision requests.