A multiweight mesh simplification algorithm for feature preservation in digital twin models

The digital twin production line recreates the real production line scene, involving vast amounts of data and resulting in high storage and computation costs, while simplifying the line models reduces these burdens, making storage, computation, and rendering more efficient. However, most simplification algorithms apply a uniform reduction across the model. As a result, feature information may be lost when the production line model, containing regions with multiple features and local features, is simplified. In this study, a multi-weight mesh simplification algorithm is proposed for feature preservation in digital twin models. Initially, spectral clustering is utilized to segment the model into feature regions and assign curvature weights, where the spectral cluster graph is constructed by an adjacency matrix and importing feature decomposition into GPU to reduce the time cost. Subsequently, weights are mapped based on the area and regularity of the facets constituting local features. Finally, these multiple weights formulate an improved cost function for mesh simplification. Experimental results demonstrate that the algorithm achieves a more than tenfold increase in segmentation efficiency and reduces simplification errors by 20.74% and 16.68% compared to the quadric error metric (QEM) and regional hierarchical QEM (RH-QEM) algorithms, respectively, while effectively and uniformly maintaining model details.