AI Teaches Itself to Tune 3D Printers, Cuts Out the Guesswork

Getting a selective laser sintering printer to produce reliable parts has always required expert guesswork. A new paper from arXiv shows that an AI agent can handle that job itself.

The system, built around a large language model and tested on the open-source SLS4All Inova Mk1, optimized process parameters for three materials: PA12 glass-filled nylon, PA11 Onyx, and a custom PA12 blend. It ran 40 experiments total, reviewed tensile and flexural test results after each batch, and adjusted laser power, chamber temperature, and energy density on its own.

The outcome: the agent found six successful configurations across different laser power levels and matched the mechanical properties listed in the material datasheets. It did this without a human engineer fine-tuning each setting. The system also kept memory of previous builds in a PostgreSQL database, so each new round of testing built on what came before.

Why this matters: SLS is precise, but fragile. Small changes in laser speed or bed temperature can turn a strong part into a brittle one. Right now, dialing in those settings takes time and specialized knowledge. If an AI can narrow the search space from millions of possibilities down to a handful of viable options, smaller workshops and universities could run SLS without hiring a process specialist.

The researchers tested the system on nylon materials commonly used for functional prototypes and end-use parts. The approach is not limited to one printer model. The team believes the same agentic framework could adapt to other additive manufacturing systems and alloys, as long as the machine exposes control parameters and returns test data.

The full paper, "AI Agentic Selective Laser Sintering Process Optimization," is available on arXiv. It was tested on the SLS4All Inova Mk1 using PA12 GF, PA11 Onyx, and a PA12 blend.