LLM-Based System for Explainable Defect Analysis in Laser Powder Bed Fusion
A newly developed decision-support system combines structured defect knowledge with reasoning from large language models to deliver clear diagnostic and mitigation recommendations for laser powder bed fusion (LPBF), a critical additive manufacturing process. This ontology-integrated system is founded on a hierarchical knowledge base that includes 27 recognized LPBF defect types, detailing their causal connections. It allows for fuzzy natural language queries to facilitate systematic knowledge retrieval, offers literature-backed explanations for defects, and provides insights into causes and mitigation approaches based on encoded process knowledge. Additionally, a multimodal image-assessment module utilizes foundation models for descriptor-guided analysis of microscopic defect images through semantic alignment scoring, aiming to improve transparency and trust in AI-enhanced manufacturing quality control.
Key facts
- System integrates structured defect knowledge with LLM-based reasoning for LPBF defect analysis
- Knowledge base contains 27 known LPBF defect types organized hierarchically
- Supports fuzzy natural language queries for knowledge retrieval
- Provides literature-supported explanations of defects
- Offers guidance on defect causes and mitigation strategies
- Includes multimodal image-assessment module using foundation models
- Image module enables descriptor-guided interpretation of microscopic defect images
- System is designed for safety-critical manufacturing applications
Entities
—