A Diffusion-Based Unified Framework for Open-World Dynamic Wheel Recognition System Construction and Maintenance With Incomplete Data

Wheel recognition systems constitute a critical foundational process in wheel manufacturing. However, dynamic open production environments and data incompleteness pose significant challenges, including missing multipainting process data, continuously emerging novel classes, and scarce training samples for novel classes, which severely impact system construction and maintenance. To address these issues, this article proposes a unified diffusion model-based framework. First, a style and triple structure guided diffusion model is introduced to synthesize high-quality wheel images for building a complete training dataset. Second, a generative out-of-distribution (OOD) mixture novelty detection method is proposed, leveraging synthesized OOD data alongside known class data to establish precise detection boundaries. Finally, the diffusion model can be used to generate sufficient training data, while self-supervised style-consistency learning is proposed to bridge the domain gap between synthetic and real images, thereby enhancing downstream task performance in wheel recognition and model updating. Experiments demonstrate that the proposed framework outperforms existing state-of-the-art approaches across multiple downstream tasks. To the best of our knowledge, this represents the first application of the diffusion model to intelligent system construction and maintenance, showcasing the potential of AIGC technology in industrial manufacturing.