An Agent-Based Solution for Printed Part Verification in Additive Manufacturing

Additive manufacturing (3D printing) offers new design freedom for engineered parts; however, verifying whether printed parts meet the intended design remains a unique challenge. In traditional manufacturing, dimensional inspection methods are mature, repeatable, and supported by long-standing standards. In contrast, verification in 3D printing is often manual, inconsistent, and fragmented. To confirm part accuracy, engineers must consider multiple factors simultaneously, design rules (such as GD&T), process-related distortions, material behavior, measurement uncertainty, and an expanding set of process and testing standards. Because these issues involve design engineering, manufacturing process physics, materials science, metrology, and regulatory standards, part verification in 3D printing requires a true multidisciplinary approach and is difficult to execute consistently.

To address this challenge, we present an agent-based system powered by large language models (LLMs) that automates specification-aware part verification. The system uses an orchestrator that coordinates three specialized agents. The first is a design/specification agent that uses retrieval-augmented generation (RAG) to extract nominal dimensions, GD&T rules, and relevant standards. The second is a metrology agent that reads measurement data, such as CMM points or diameter values, and computes uncertainty-aware measurements using the three criteria. The third is a quality-assurance (QA) agent that compares results to tolerances and generates traceable, human-readable reports.
We demonstrate the workflow through a laser powder bed fusion (PBF) example focused on verifying hole features. The system automatically retrieves the nominal hole specification, computes uncertainty-aware diameters, and produces a transparent, traceable decision with step-by-step reasoning. We also integrate a human-in-the-loop component, allowing engineers to review the agent-generated results, provide feedback, and verify the final outcome. This interactive step helps maintain trust and supports continuous improvement of the agent-based workflow. Overall, the results show that the approach reduces manual effort, improves consistency in applying standards, and delivers explainable and auditable verification decisions.