AI Based Optimization of Architectural Designs and Mechanical Systems
Abstract
The present study explored the mediating role of AI-enabled optimization between architectural design decisions and mechanical system performance during the early design stage in building projects. This study aimed to evaluate the robustness of and construct validity with this framework. We apply the estimation method of difference-in-differences, where retrofit program rollouts producing exogenous variation (in the event-study analysis) and energy spending on non-HVAC categories directly not affected by the optimization (in the placebo analysis) serve as comparison groups. Drawing on a multi-level SEM of a project consortium of over 120 members, the paper first addresses aspects of the intervention that could be seen as beneficial and/or disruptive—particularly in terms of coordination costs and model transparency. Our results suggest strong positive short-run effects of the optimization on project teams’ simulation volume, and moderate effects on annual energy use and thermal comfort indices. These results correspond to the regression-estimated effects on utility spending on electricity. Yet while perceived by practitioners as a relatively routine and low-friction aspect of their professional lives, the research also points to instances of the deployment that constituted burdensome, opaque and/or exclusionary forms of stakeholder engagement. The institutionalization of multi-criteria priorities through AHP may lead to clearer accountability for architects and engineers, contributing towards governance templates that will scale. In these settings, the analysis highlights tensions between what appears to ‘work’ for teams in the short term and likely long-run implications that these practices might have for equity and sustainability of outcomes.