Answers are grounded in this paper and cite their sources.
Validation study demonstrates accurate manufacturing feasibility prediction in 3D mechanical designs, indicating effective early-stage screening through hybrid feature modeling.
Early-stage feasibility assessment of 3D mechanical designs remains a critical challenge in manufacturing, where inefficient screening increases development costs and delays decision-making. Conventional approaches rely on simplified scalar constraints or computationally intensive simulations, poorly suited to iterative design. This study proposes a machine learning framework integrating scalar design parameters with geometric features extracted from three-dimensional models for feasibility prediction. A balanced dataset of 654 mechanical design instances, obtained from an industrial metal-manufacturing partner and equally divided into feasible and infeasible cases, is used to develop and evaluate the framework. Each design was labelled empirically from its actual fabrication outcome, ensuring feasibility reflects observed manufacturing results rather than rule-based or simulated criteria. Scalar variables are standardized, while geometric features such as volume, aspect ratio, and bounding box dimensions are extracted and combined into a unified representation refined through statistical analysis and feature selection. Multiple classifiers are trained using stratified cross-validation, and generalization is assessed using a group-wise protocol that withholds entire design groups from training. The framework attains approximately 81% internal accuracy and around 84% accuracy on previously unseen design groups (ROC-AUC approximately 0.89). An ablation analysis shows scalar constraints account for most of the predictive signal, while geometric descriptors contribute marginally for the product families considered, characterizing feature contributions empirically rather than assuming a benefit. Calibrated probabilistic outputs feed a three-zone decision policy separating automatic decisions from cases requiring manual review. The framework suits early-stage design screening under controlled conditions and offers a rigorously validated, interpretable approach for data-driven feasibility evaluation.
No takes yet. Share an insight, caveat, or question.
Fahim et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: