About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Accelerating Innovation in Materials and Manufacturing
|
| Presentation Title |
Agentic Systems Design of an Open-Source Powder Doser for L-PBF Feedstock Research: CAD, PCB, and Firmware |
| Author(s) |
Sam Charles, William Mulberry, Luke Winters, Sterling Baird |
| On-Site Speaker (Planned) |
Luke Winters |
| Abstract Scope |
Agentic AI and generative design tools promise faster laboratory hardware development, but their deployable capabilities across CAD, electronics, and firmware remain poorly characterized. We report an engineer-led assessment spanning the systems-design stack of an open-source powder doser for laser powder bed fusion feedstock research. Using multiple generative-CAD and agent-coding platforms, we produced parametric CadQuery and OpenSCAD models, KiCad schematics, and MicroPython firmware for multi-actuator control with load-cell feedback. Fully AI-generated designs were not deployment-ready—outputs carried geometric errors and hallucinated features—so engineers defined geometry, dimensions, hand sketches, and manufacturing constraints; AI modeled the parts; engineers reviewed, printed, and iterated. Higher-specificity prompts improved usable output, yet weak spatial reasoning remained a recurring limitation across many correction cycles. This open-source route can broaden access to powder-handling instrumentation and reduce dependence on proprietary development pipelines. We present prototypes, workflow tradeoffs, and guidance for AI-assisted development of materials-lab hardware. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Process Technology, Computational Materials Science & Engineering, Machine Learning |