The growing complexity of infrastructure systems calls for an inspection program that can maximize the value of inspection for maintenance and rehabilitation decision support by dynamically adapting to the systems’ uncertain and heterogeneous deterioration behaviors. While current inspection programs are mostly periodic and do not satisfy the need, recent risk-informed inspection optimization frameworks either depend on predefined rules or are limited to individual assets or components, restricting its flexibility and scalability across heterogeneous infrastructure systems. To bridge the gap, this study proposes a data-driven, risk-informed framework for developing dynamic inspection policies of a large-scale bridge inventory. The three-stage framework integrates an inventory-level stochastic deterioration model, a stochastic sequential decision process for joint dynamic inspection and replacement optimization, and a clustering-based machine learning process to develop a cross-asset transferable inspection program. Two competing non-periodic inspection policies — time- and condition-based — are formulated and compared in terms of their long-term life-cycle economic efficiency. Results confirm the economic advantage of condition-based inspection strategy. Moreover, it is found that the optimal inspection policy can be compactly represented by a one-parameter truncated linear function. Using the proposed framework, the provincial bridge inventory of Ontario, Canada is examined as a case study, for which an inventory-level dynamic bridge inspection program is developed.
Chen et al. (2026) studied this question.