Initial release of the few-shot bearing fault diagnosis framework using multimodal LLMs and prototypical networks.
Features:
- Support for multiple MLLMs: GPT-4o, GPT-5.1, Claude 4.5 Haiku/Sonnet, LLaVA-1.5-7B
- Prototypical Networks baseline with ResNet-50 and Swin Transformer V2-T
- 1-shot, 5-shot, and 10-shot evaluation
- Automated metrics computation with confidence intervals
- CWT image preprocessing for vibration signal analysis
Requirements:
- Python 3.8+
- OpenAI/Anthropic API keys for cloud models
- User-provided bearing vibration datasets (CWT images)
See README.md for installation and usage instructions.