To address the challenges of lengthy design cycles and high verification costs in the casting industry, which arise from insufficient design experience during the research and development of complex new parts, this study proposes a similarity measurement method for casting parts tailored to intelligent design assistance. An adaptive multifeature fusion framework based on adaptive weight allocation was constructed, which enables efficient comparison with production-verified mature cases in the historical part database and provides decision support for new part design. Leveraging Python, the method implements multi-dimensional feature fusion, incorporating methods including feature preprocessing, 3D point cloud sampling, part spatial slicing and KD-tree construction, as well as cosine distance and surface normal vector matching for similarity evaluation, significantly enhancing the effectiveness and efficiency of similarity calculations. Experimental results demonstrate that when adaptive weights control the proportion of each feature extraction, the optimal part similarity reaches 81.25% for the example parts in this study. This occurs when the weighted values of cosine similarity, KD-tree similarity, slicing similarity, and normal vectors are set to 0.2, 0.15, 0.4, and 0.25 respectively—A performance that significantly outperforms any single-feature measurement. The proposed algorithm has been integrated into 3D design software and validated through case studies, proving its capability to rapidly and accurately retrieve and recommend similar parts and their mold structures. This provides reliable references for casting design, thereby reducing redundant design processes, shortening the research and development cycle, and lowering verification costs.