Physics-informed multi-objective placement of indoor multi-modal sensor nodes using a two-dimensional LBM-based simulation framework

This study presents a physics-informed multi-objective optimization (PI-MOO) method for placing integrated indoor sensor nodes. A two-dimensional lattice Boltzmann method (LBM) model was built on the breathing-zone plane (z = 1.2 m) of an educational building floor and used to generate reference fields for temperature and pollutant concentration variance under summer forced-convection and transitional-season natural-ventilation conditions. NSGA-II optimized three objectives: temperature-field reconstruction error (RMSE), line-of-sight (LoS) coverage, and variance-weighted gas-risk coverage. The Kneedle algorithm selected N = 31 as the cost-effective sensor budget. At this budget, PI-MOO reduced RMSE by 30.3% relative to uniform placement and by 23.6% relative to the virtual-force algorithm; gas-risk coverage increased by 24.6% and 7.7%, respectively. In a Monte Carlo node-failure test, PI-MOO kept the best absolute performance after five failed nodes. The 2D LBM fields are used here as reference fields for comparing layouts. Validation against three-dimensional simulations or field measurements remains future work.