BESS Component Detection and Digital Twin Validation Dataset

This record contains the YOLO-format image dataset used for Stage I component detection in an AI-driven Digital Twin framework for Battery Energy Storage System (BESS) validation and troubleshooting. Version 1.1.0 contains 15,099 images and 15,099 matching label files, divided into 14,436 training images and 663 validation images. The dataset uses 62 configured class identifiers, of which 61 are represented by at least one annotation in the complete dataset and 48 are represented in the validation split. It contains 98,570 YOLO annotation rows: 82,094 bounding-box rows and 16,476 polygon rows accepted by Ultralytics. No separate held-out YOLO test split was retained. The record also includes source and licence manifests, file-level provenance, cross-split integrity information, model and training records, representative visual evidence, and supporting artifacts for the reported Stage I, Stage II, Stage III, controlled-scenario, end-to-end, and live LLM evaluations. The dataset combines attributed third-party source images with project-generated synthetic compositions. Source-level provenance and licensing information are documented in SOURCE_MANIFEST.csv and LICENSES.md.