Rule-Based Geometric Placement and Optimization-Ready UV Parameterization for Generative Facility Layout Design in BIM Environments: A Proof-of-Concept Preparation Methodology

Generative design (GD) in Building Information Modelling (BIM) frequently yields infeasible or uninformative alternatives. Within the scope of this study, we examine how GD outcomes relate to preparation before optimization: how placement rules are encoded and how the search space is parameterized so that an evolutionary engine can exploit a stable genotype–phenotype mapping. We present a three-module preparation framework for facility layout GD in BIM, implemented in Dynamo with the NSGA-II-based Revit GD engine. Module A abstracts spaces as rectangular units and maintains a list of valid, unoccupied candidate locations through surface-based containment tests and Boolean-mask filtering. Module B enumerates discrete rotation and adjacency scenarios and prunes true collisions with a tolerance-controlled intersection test (16 of 32 raw scenarios rejected in a worked example). Module C replaces shuffle-based randomization, which re-maps a genotype to a different layout at every evaluation, with a deterministic, locality-preserving normalized-coordinate (UV) encoding. Modules A and B are demonstrated through worked geometric examples and Module C through a synthetic proof-of-concept run with three space groups and piping-length objectives on an irregular site, in which later-generation layouts showed reduced routing values and clustering of connected spaces, consistent with the representational argument. An additional run on the same case connects the three module operations under a restricted rotation–adjacency construction. The framework is offered as a documented preparation protocol bounded by a single illustrative case.