Linking microstructure informatics with characterization knowledge in additively manufactured composites through customized and hybrid vision-language representations for automated qualification

Rapid and reliable qualification of advanced materials remains a bottleneck in industrial manufacturing, particularly for heterogeneous structures produced via non-conventional additive manufacturing processes. This study introduces a novel framework that links microstructure informatics with a range of expert characterization knowledge using customized and hybrid vision-language representations (VLRs). By integrating deep semantic segmentation with pre-trained multi-modal models (CLIP and FLAVA), we encode both visual microstructural data and textual expert assessments into shared representations. To overcome limitations in general-purpose embeddings, we developed a customized similarity-based representation that incorporates both positive and negative references from expert-annotated images and their associated textual descriptions. This allowed zero-shot classification of previously unseen microstructures through a net similarity scoring approach. Validation on an additively manufactured metal matrix composite (MMC) dataset demonstrated the framework’s ability to distinguish between acceptable and defective samples across a range of characterization criteria with up to 80% top-5 retrieval accuracy. Comparative analysis revealed that FLAVA model offers higher visual sensitivity and penalized weak similarities with score differences as large as 0.17 relative to CLIP. However, FLAVA’s text encoder exhibited sharp drops in similarity for paraphrased expert descriptions (falling below 0.20), whereas CLIP maintained more stable alignment with textual criteria (0.29-0.36). Z-score normalization adjusted raw unimodal and cross-modal similarity scores based on their local dataset-driven distributions, enabling more effective alignment and classification in the hybrid vision-language framework. The standardized scores provided strong binary classification results across three categories (82% for distribution, 90% for dilution, and 82% for reinforcement). The proposed method enhanced traceability and interpretability in qualification pipelines via human-in-the-loop decision-making without task-specific model retraining. By advancing semantic interoperability between raw data and expert knowledge, this work contributes toward scalable and domain-adaptable qualification strategies in engineering informatics.