DART: Vision-Language Model for Rope Condition Monitoring
A new foundation model for AI, named DART (Damage Assessment via Rope Transformer), has been created to thoroughly monitor the condition of synthetic fibre ropes (SFRs) utilized in offshore, maritime, and industrial applications. Unlike conventional classifiers, DART offers ongoing severity assessments, maintenance suggestions, anomaly alerts, deterioration forecasts, and automated reports based on a single inspection image. This model enhances the Joint-Embedding Predictive Architecture (JEPA) by integrating a Vision Transformer (ViT-H/14) with Llama-3.2-3B-Instruct through a Severity-Conditioned Cross-Modal Fusion (SC-CMF) module. Key innovations include HD-MASK, a saliency-guided masking technique; learnable severity gates for each class; and a cohesive multi-task architecture. The findings are available on arXiv with the identifier 2605.04943.
Key facts
- DART stands for Damage Assessment via Rope Transformer
- Model addresses full rope inspection workflow
- Uses Vision Transformer (ViT-H/14) and Llama-3.2-3B-Instruct
- Includes Severity-Conditioned Cross-Modal Fusion (SC-CMF) module
- HD-MASK is a saliency-guided masking strategy
- Provides severity estimates, maintenance recommendations, anomaly flags, deterioration timelines, and automated reports
- Published on arXiv with ID 2605.04943
- Designed for synthetic fibre ropes (SFRs) in offshore, maritime, and industrial settings
Entities
Institutions
- arXiv