Abstract: This paper addresses the challenge of efficient management for massive parametric 3D model data assets in the manufacturing domain by proposing a novel 3D model analysis method based on feature extraction and multi-view clustering. Through decomposition of 3D models, we extract four categories of features (geometric, topological, surface, and medial surface) to construct a comprehensive multi-view representation system. To overcome the limitations of conventional clustering methods that are prone to local optima and sensitive to noise, we propose a unified latent space-based multi-view clustering approach. Our method learns latent low-dimensional embeddings from multi-view data, constructs relational graphs in a consistent latent space, and optimizes the graph structure using robust estimators with trace constraints before applying spectral clustering. Experimental results on the TraceParts standard 3D model dataset demonstrate that our method significantly outperforms the state-of-the-art MCLES (Multi-view Clustering in Latent Embedding Space) algorithm, achieving improvements of 5.5% in ACC, 8.0% in NMI, 9.2% in F-score, and 13.8% in Precision.