Raw Yarn-Breakage Image Dataset from Robot-Assisted Ring-Spinning Inspection (Textile Manufacturing, 2024)

The class distribution is: 000g, 71 images; 001g, 134 images; 011g, 300 images; 100g, 120 images; 101g, 416 images; 110g, 404 images; and 111g, 2388 images. The Raw images are original unprocessed image records. No preprocessing, enhancement, filtering, resizing, normalization, or data augmentation has been applied to the Raw images.Data Collection & ProcessingData were collected in an industrial textile factory under actual ring-spinning production conditions, where the yarn production-line speed could reach approximately 3000 m/min.

The mobile inspection platform performed fixed-path robot-assisted stop-and-capture inspection along the production line at a patrol speed of 0.5 m/s. Image acquisition was conducted at planned stop locations spaced at 1.2 m intervals, and each image includes the left-middle-right spindle region under a fixed imaging configuration.The dataset was collected across three acquisition dates and five major acquisition sessions, covering different acquisition times and workshop background regions, thereby increasing image diversity and supporting robustness evaluation.

Each Raw BMP image is treated as an image-level timestamped inspection record for machine-learning use. Based on the timestamped acquisition records, the 3833 Raw BMP images can be further organized into 162 timestamp-derived acquisition groups, providing acquisition-structure-level traceability under the fixed-path robot-assisted inspection protocol.Metadata DetailsA shared common_metadata/ folder is provided for both the Raw and Preprocessed records.

It includes dataset structure, class definitions, acquisition protocol, preprocessing description, image-level temporal metadata, Raw/Preprocessed pairing information, and quality-control documentation.This dataset supports machine-vision research and textile defect inspection for industrial yarn-breakage classification and detection, including whole-image classification, ROI-based analysis, and further annotation-based object-detection studies.