United States Department of Transportation
i
AI-enabled Fiscally Constrained Lifecycle Asset Management for Infrastructure Systems
-
2021-11-30
-
[PDF-1.55 MB]
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Corporate Publisher:
-
Abstract:Accurate evaluation and prediction of infrastructure components performance and determination of optimal maintenance and inspection actions for any infrastructure element throughout its lifetime are essential parts of an effective infrastructure asset management framework. The goals of this project were to develop Artificial lntelligence (AI)-enabled solutions, providing infrastructure condition assessment and prediction models, as well as algorithms able to directly suggest optimal maintenance and inspection decisions for multi-component infrastructure systems over long planning horizons. This report develops a prediction and decision-making framework for inspecting and maintaining deteriorating systems with incomplete information and constraints. In doing so, a Partially Observable Markov Decision Processes (POMDPs) approach is used, with an original deep reinforcement learning formulation. Thus, a Deep Decentralized Multiagent Actor-Critic (DDMAC) architecture is devised and manages to successfully tackle numerous challenges imposed by this stochastic control problem. Various constraints are also effectively incorporated in this framework. Further, a deterioration model for bridge decks using Random Survival Forest is developed. The results suggest that AI methods can achieve high accuracy in predicting the deterioration pattern of bridge decks, which is an important input into the stochastic optimal control framework. However, while AI methods may be preferred for prediction, because it is difficult to interpret the impacts of different variables on deterioration, traditional stochastic methods can be more powerful for construction or design purposes.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
