Risk-based Inspection Planning for Reinforced Concrete Bridges

Bridges are vital components of transportation systems. Bridge decks deteriorate faster than other bridge components due to their direct exposure to traffic loads, harsh environmental conditions, and de-icing chemicals. Inspecting these decks allows for the timely detection of defects, providing data for actionable maintenance decisions, and efficiently allocating resources. Accurate and cost-effective inspection faces many challenges, ranging from standardization of data collection, risk consideration, condition-based inspection frequency, and efficient utilization of Non-destructive Evaluation (NDE) methods. This research presents an automated, risk-based inspection planning method for reinforced concrete bridges, focusing on bridge decks. This method integrates five main automated modules: (1) a data integration and standardization module, developed to automatically capture, structure, and integrate bridge data from diverse sources, including bridge inspection reports. It also standardizes inspection data into a ready-to-use format that supports direct, subsequent utilization in condition assessment and predictive modeling. It employs web scraping and rule-based data extraction techniques, in addition to text mining, natural language processing, and state-of-the-art large language models. (2) an automated, data-driven condition assessment module, developed to enable quantitative condition assessments of bridge decks using Bayesian belief networks. It makes use of the data generated in the first module. (3) a risk-of-failure module, developed to incorporate both the probability of failure and the consequences of failure, employing advanced ensemble machine learning and deep learning techniques, as well as a fuzzy inference system. (4) an NDE effectiveness module developed to assess the performance of NDE methods based on their respective accuracy, precision, speed, cost, and ease of use. It also evaluates the effectiveness of integrating multiple of these methods to optimize inspection quality and reliability. (5) a dynamic multi-objective optimization module for inspection planning, developed to strike a balance between the structural risk of failure, inspection effectiveness, and inspection costs (both direct costs and impact costs of inspections) under budgetary and operational constraints. This module enables data-driven decision-making to identify which bridges require inspections, the optimal timing for these inspections, and the most effective NDE method for each case. The developed modules have been tested and validated utilizing inspection data of a set of bridges in Québec, Canada. The data comprises 4,119 bridge structures and 2,255 bridge inspection reports spanning a five-year period from 2018 to 2022. The developed modules demonstrated good performance with significant potential to improve bridge asset management practices. For instance, the defect-based condition assessment module achieved an accuracy of 94.6%, precision of 95.0%, recall of 94.7%, and F1 score of 94.1%. Inspection data was also transformed into a ready-to-use format with an accuracy of 98.79% for detecting and characterizing rebar corrosion, 99.09% for concrete delamination, and 98.64% for cracking, scaling, and spalling of concrete. The developed probability of failure model, as a part of the risk-of-failure module, achieved a mean accuracy of 98.84% with a 0.13% standard deviation. The developed inspection schedules effectively adopted inspection methods based on structural conditions, risk levels, and trade-offs among structural risk of failure, inspection costs, and inspection effectiveness. According to the developed schedules, high-risk bridges should receive more frequent and effective inspections earlier in the planning horizon, while bridges with lower risk need fewer, less intensive inspections. The developed method has the potential to improve inspection effectiveness, reduce costs, rationalize intervention planning (including associated budget and resource allocation), and enable broader, effective utilization of NDE methods.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.