AgentsCAD Automated Design for Manufacturing of FDM Parts via Multi-Agent LLM Reasoning and Geometric Feature Recognition

Jul 16, 2026

Fused Deposition Modeling parts often need Design for Additive Manufacturing updates to improve printability, structural integrity, and reduce post-processing, yet slicers can flag defects without changing the underlying geometry. AgentsCAD automates targeted DFAM by parsing STEP B-Rep data, detecting overhangs beyond a 45° threshold, building a face-adjacency topology graph, and using multi-agent LLM reasoning to recommend corrective CAD operations. A vision-language verifier checks geometric integrity from rendered views, producing a modified STEP file and a readable report that closes the geometry-to-language modification loop.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages28

Reading Time1 h 11 min

Abstract

Introduction

arXiv :2607 .02448v2 [ cs .MA] 7 Jul 2026 AgentsCAD: Automated Design for Manufacturing of FDM Parts via Multi-Agent LLM Reasoning and Geometric Feature Recognition Emmanuel George 1 , Christopher Keefe2 , Peter Pak3 , and Amir Barati Farimani ∗ Department of Mechanical Engineering, Carnegie Mellon University, Pittsburgh, PA, USA E-mail: [barati@cmu.edu](barati@cmu.edu) Abstract Parts manufactured with Fused Deposition Modeling (FDM) often require Design for Additive Manufacturing (DFAM) modifications to ensure printability, structural integrity, and reduced post-processing. Current slicers identify defects such as steep overhangs but are unable to modify the underlying geometry. This work presents AgentsCAD, a multi-agent system that bridges raw boundary-representation (B-Rep) geometry and Large Language Model (LLM) reasoning to automate targeted DFM. The workflow begins by parsing a STEP file. The agentic system detects overhangs above a 45°threshold, constructs a face-adjacency topology graph, and optionally injects semantic feature labels from a GraphSAGE model trained on MFCAD++ (59,665 parts), before dispatching a Claude Sonnet design-reasoning agent that recommends reorientations, fillets, chamfers, and similar modifications. A GPT-4o vision-language verifier inspects rendered views to confirm geometric integrity. Outputs include a modified STEP file and a human-readable report. A test case on a birdhouse model demonstrates that the system correctly diagnoses overhangs, selects appropriate defect mitigation strategies, and proposes physically valid corrections, partially solving the geometry-to-language translation problem central to LLM-driven CAD modification. 1 Introduction Fused Deposition Modeling (FDM) is an additive manufacturing process which builds parts layer by layer through depositing material along a sequence of two-dimensional crosssections. 1–3 Since each new layer must rest upon the one beneath it, downward-facing surfaces tilted beyond roughly 45° from vertical (overhangs) are difficult to print reliably without supports or geometric modification. Design for Additive Manufacturing (DFAM) 3,4 addresses these limitations by reshaping geometry to improve dimensional accuracy, mechanical strength, and print efficiency while minimizing post-processing. Modern slicing software can flag problematic regions and generate support structures, but be corrective decisions must be made by the engineer, who must iterate manually between CAD and slicer platforms until the part is suitable for printing. A small but rapidly growing body of work has begun applying language and vision-language models at the CAD interface, organized in recent surveys along generation, modification, and analysis axes. 5,6 On the generation side, Text2CAD produces sequential CAD designs from natural-language prompts spanning beginner-to-expert detail, 7 Query2CAD generates parametric models directly from natural-language queries using FreeCAD, 8 and CADSmith 9 utilizes CADQuery based Python bindings to generate parametric models. ChatCAD extends this idea to multimodal LLM-guided CAD drawing restoration in zero-shot settings, 10 and CADCodeVerify uses vision-language models to validate generated CAD code. 11 In additive manufacturing specifically, large language models have been applied to the domain of defect prediction in various build monitoring tasks, 2,3,12 complementing real-time vision-based defectdetection systems for FFF printing. 13 These systems share a common pattern in that language or images map to new geometry or to defect classes, but no system to the authors’ knowledge modifies an existing part’s underlying B-Rep for more favorable manufacturing conditions. Complementary literature on Automatic Feature Recognition (AFR) provides the geometric understanding needed to fill that gap. 14 Early work used hand-engineered rules and shallow neural networks to map manufacturing features from CAD geometry. 15,16

FAQs

What problem does AgentsCAD address in FDM additive manufacturing?

AgentsCAD targets the gap where current tools can detect printing issues (like steep overhangs) but cannot modify the underlying CAD geometry to make parts manufacturable.

How does the workflow identify manufacturability issues before proposing edits?

The system parses STEP files, detects overhangs using a 45° threshold, and constructs a face-adjacency topology graph that can be enriched with semantic feature labels from a GraphSAGE model.

How are proposed CAD modifications validated and what outputs are produced?

A GPT-4o vision-language verifier inspects rendered views to confirm geometric integrity. The system outputs a modified STEP file plus a human-readable report detailing the recommended changes.

Jul 16, 2026

Upload by Ava Thompson