Toward a Trustworthy AI-Based Decision Support for Bridge Asset Management
TL;DR
Compared with traditional reactive, condition-based decision policies, the proposed AI-based policies are found to offer more proactive and adaptive decisions and to demonstrate superior performance in minimizing the life cycle costs.
Abstract
This study develops a novel artificial intelligence (AI)-based methodology to enable risk-based, long-term-focused decision support for bridge asset management. This methodology features three key modules: (1) a bridge condition simulation module integrating Markovian state transition dynamics, (2) a lifecycle cost analysis module to holistically quantify and aggregate the direct and indirect costs incurred due to bridge condition deterioration and intervention actions over a prolonged planning horizon, and (3) an AI-based decision-support module for optimal sequential maintenance actions at individual bridge level. The proposed methodology can flexibly accommodate the decision needs of different individual bridges within a bridge portfolio under various sources of uncertainty in near real time. Compared with traditional reactive, condition-based decision policies, the proposed AI-based policies are found to offer more proactive and adaptive decisions and to demonstrate superior performance in minimizing the life cycle costs. To further enhance risk-averse stakeholders’ trustworthiness in the AI-based decision policy, practical action constraints and decision reasoning are introduced, and the effect of action override is also investigated.
