Artificial intelligence aided resilient and sustainable water infrastructure systems

Water infrastructure systems are a pillar of communities as they supply water to the public in both normal and emergency conditions. Due to climate change and aging, the water infrastructure systems are facing more uncertainty and damages. Unexpected failures and inefficient management will significantly increase the tension among existing infrastructure, the environment, and financial burdens. They will also amplify the risks and losses in case of hazards. Therefore, there is an urgent need to improve the resilience and sustainability of existing water infrastructure systems. Recently, with the rapid development of data acquisition systems and Internet of Things, a large amount of data is continuously generated and collected from water infrastructure systems. A complete mining of these data is essential to enriching the current understanding of water infrastructure failures and to supporting optimal decisions in infrastructure management. However, traditional methods are limited due to the large number of samples and complex relationships in the dataset. The recent progress in Artificial Intelligence (AI), especially in machine learning (ML), has shown promising solutions for analyzing such large datasets. The AI techniques are more robust and computing efficient in learning complex relationships from large datasets, as has been demonstrated in multiple applications. As a result, there is a growing interest in using AI to improve the resilience and sustainability of water infrastructure systems. In this study, we leveraged the power of AI, simulation models, and large historical datasets. This study focused on three main components of water infrastructure management: water system leak detection, water pipe failure prediction, and decision-making in disaster management. Specifically, a clustering-then-localization semi-supervised learning (CtL-SSL) algorithm was proposed for the water system leak detection. The results showed that the proposed algorithm can achieve higher leak detection accuracy and localization accuracy without a need for extensive labeled dataset. This study also proposed a comprehensive framework for using AI for pipe and water system failure prediction. The results from the AI models were interpreted and their uncertainties were quantified in order to increase their trustworthiness and reliability. Lastly, this study aimed to improve the resilience of water infrastructure systems by facilitating optimal decisions after the occurrence of hazards. A graph convolutional neural network-based reinforcement learning decision-making tool was proposed. The results indicated the proposed algorithm achieved a higher resilience index than traditional decision-making methods. Overall, AI has emerged as a powerful technology and shown potential for intelligent infrastructure systems? management. The methods proposed in this study can be used for prolonging the service time of water pipes and making optimal decisions in water system management. Therefore, applying this study in water infrastructure system can reliably improve their resilience and sustainability.