GMCAD: an original Synthetic Dataset of 2D Designs along their Geometrical and Mechanical Conditions

We build an original synthetic dataset of 2D mechanical designs alongside their and geometric constraints, GMCAD. Such a allows training Deep Learning (DL) models for Design Additive Manufacturing (DfAM) to incorporate control Computer-Aided-Design (CAD) features with performance. Geometric AM constraints are often complex describe, depending on applications, processes, materials. They lack explicit mathematical descriptions, belong exclusively the CAD world, hardly can be integrated into design, hampering design freedom. DL have recently emerged as potential reconcile both Computer Aided-Engineering (CAE) worlds. derive data-driven rules over designs, allowing fine-grained geometry during phase, contrary conventional CAD-to-CAE sequential approach. models, however, need high-quality labeled data, merging CAE aspects is challenging they rely different formats, rules, tools. GMCAD solves this issue following these building steps. (i) Building DL-mechanical conditions predictor from generated by density-gradient-based Topology Optimization method (TO); AM-synergetic generation tool. (ii) Creating inspired TO-based designs. (iii) Predicting CADs using conditions. Last, we evaluate performance GMCAD’s statistics features. Designs show significant influence minor changes, explaining intricate task conforming functionality constraints. Consequently, having advantageous train generate accounting all simultaneously, without time-consuming trial error techniques. could enhance DfAM go beyond AM; also other fields automatic reconstruction, reverse engineering, isogeometric paves way multi-objective controllable generation.