A data-driven framework for causal modeling of design change propagation in civil infrastructure

Design changes are a major source of cost overruns and delays in civil infrastructure projects, and yet the mechanisms by which such changes propagate across interdependent components remain insufficiently understood. Existing approaches to modeling component dependencies often depend on information sources that are inherently subjective, incomplete, or difficult to specify exhaustively in complex civil infrastructure projects. In addition, inferred dependency structures are seldom grounded in formal statistical inference, limiting confidence in their robustness and interpretability. This study develops a data-driven framework that infers change-propagation structures directly from geometric differences between initial and final design models. The project space is partitioned into spatial grids to construct a binary change indicator matrix (CIM). Intercategory dependencies are inferred using constraint-based causal discovery, and propagation probabilities are estimated via maximum likelihood under a Noisy-OR formulation. Direct and indirect effects are aggregated into a propagation probability matrix (PPM). The framework then is applied to a large-scale expressway construction project in Japan using thousands of design drawings and 159 change-order (CO) records. The effects of spatial grid size and geometric change-detection threshold are examined based on their ability to reproduce documented intercategory relations. Spatial discretization is found to play a dominant role in recovering these relations, whereas the influence of geometric tolerance is comparatively minor. Under the selected parameter setting, all documented intercategory relations are reproduced in the inferred network. The resulting PPM exhibits a structured and hierarchically asymmetric pattern: Propagation is strongest among primary structural components, substantial within secondary components, and consistently weak from secondary to primary elements. These findings demonstrate that propagation networks inferred through formal statistical procedures can capture empirically observed change relations, providing a scalable and data-grounded basis for analyzing change propagation in civil infrastructure systems.