Inferring global and detailed structural building variables from prior probabilities, historical structural design and Bayesian Networks

Abstract The lack of data on existing buildings inhibits circular economy strategies, such as reuse. To overcome this issue, the current study infers unknowns using Bayesian Networks (BNs), which are directed acyclic graphs of probabilistic variables. The study proposes two types of BNs to infer probabilities of variables of existing buildings. The first BN is based on conditional probability tables and associations between global building variables. The second builds on this with detailed structural load variables via engineering equations from historical structural design codes. The BNs generated probabilistic estimates which reflect uncertainty in the input data. With increased evidence, the BNs' estimates were updated, thereby reducing uncertainty of inferred building variables. Urban planners can use the tool to estimate building variables without physically measuring existing buildings, thereby enabling circular construction planning. Future studies may expand the BN to include more structural design equations to infer additional structural building variables.