Techno Press
You logged in as Techno Press

Smart Structures and Systems
  Volume 30, Number 6, December 2022 , pages 571-582

Deep reinforcement learning for optimal life-cycle management of deteriorating regional bridges using double-deep Q-networks
Xiaoming Lei and You Dong

    Optimal life-cycle management is a challenging issue for deteriorating regional bridges. Due to the complexity of regional bridge structural conditions and a large number of inspection and maintenance actions, decision-makers generally choose traditional passive management strategies. They are less efficiency and cost-effectiveness. This paper suggests a deep reinforcement learning framework employing double-deep Q-networks (DDQNs) to improve the life-cycle management of deteriorating regional bridges to tackle these problems. It could produce optimal maintenance plans considering restrictions to maximize maintenance cost-effectiveness to the greatest extent possible. DDQNs method could handle the problem of the overestimation of Q-values in the Nature DQNs. This study also identifies regional bridge deterioration characteristics and the consequence of scheduled maintenance from years of inspection data. To validate the proposed method, a case study containing hundreds of bridges is used to develop optimal life-cycle management strategies. The optimization solutions recommend fewer replacement actions and prefer preventative repair actions when bridges are damaged or are expected to be damaged. By employing the optimal life-cycle regional maintenance strategies, the conditions of bridges can be controlled to a good level. Compared to the nature DQNs, DDQNs offer an optimized scheme containing fewer low-condition bridges and a more costeffective life-cycle management plan.
Key Words
    condition assessment; deteriorating structures; life-cycle management; regional bridges; reinforcement learning
Department of Civil and Environmental Engineering, The Hong Kong Polytechnic University, Hong Kong, China.

Techno-Press: Publishers of international journals and conference proceedings.       Copyright © 2023 Techno Press
P.O. Box 33, Yuseong, Daejeon 305-600 Korea, Tel: +82-42-828-7996, Fax : +82-42-828-7997, Email: