Circular economy strategies, such as reusing load-bearing elements from obsolete buildings, can reduce the construction industry's environmental impact. However, limited data on building composition often hinders implementation. Existing methods for estimating building composition are often deterministic, subjective, or lack detail at the structural element level. This study proposes a probabilistic modelling approach using Bayesian Networks (BNs) to improve estimates of structural element dimensions and material quantities in residential timber buildings. The model integrates building variables, from twenty thousand buildings for derivation of prior modelling assumptions, with historical Norwegian and Eurocode structural design equations. When applied to four real buildings, the model delivers uncertainty-aware estimates of beam dimensions and material quantities. Our BN substantially expands the scope of prior studies regarding both the number of variables and the size of the datasets considered, enabling more robust predictions at the level of individual structural elements. Results show that BNs can estimate structural variables even with incomplete data, supporting early-stage urban planning in circular construction. Urban planners can use this model to quantify variability in structural properties prior to inspections. The element-level detail enhances estimates of future material availability and greenhouse gas emissions, improving sustainability strategies for building stocks. • New method to infer material quantities in buildings with incomplete/uncertain data. • Prior probabilities from dataset of general building data for ≈20 000 buildings. • Structural equations from historical design codes integrated with this building data. • Bayesian network with 39 variables infers beam dimensions in existing buildings.
Mutale et al. (Tue,) studied this question.