Deep learning for critical infrastructure resilience
Ensuring the resiliency of critical infrastructures is essential in modern society, but much of the deployed infrastructure has yet to fully leverage modern technical developments. This paper intersects two unique fields-deep learning and critical infrastructure protection-and illustrates how deep learning can improve resiliency within the electricity sector. Machine vision is the combination of machine intelligence, or computer systems automatically learning patterns from exemplar data, and image analysis, or objects of interest being automatically segmented and identified from video image data. This technology has the potential to automate threat assessments in the context of securing critical infrastructures. Rather than traditional reactionary approaches, we present here a method of leveraging deep learning for the detection of threats to critical infrastructures before failures occur. This paper discusses the state-of-the-art in deep learning for creating machine vision systems, and the concepts are applied to increase the resiliency of critical infrastructures. The intersection between machine vision and critical infrastructures is discussed, as are key benefits and challenges of invoking such an approach, and examples within several fields of critical infrastructures are presented. Automated inspection of the power infrastructure using vehicle-mounted video acquisition equipment is explored, and a proof-of-concept implementation of a deep convolutional neural network is developed, achieving 95.5% accuracy in distinguishing power-related infrastructures within images largely typical of rural settings. These preliminary results show promise in the application of deep learning and machine vision to protecting critical infrastructures through preventative maintenance.
|Journal||Journal of Infrastructure Systems|
Dick, K. (Kevin), Russell, L. (Luke), Dosso, Y.S. (Yasmina Souley), Kwamena, F. (Felix), & Green, J. (2019). Deep learning for critical infrastructure resilience. Journal of Infrastructure Systems, 25(2). doi:10.1061/(ASCE)IS.1943-555X.0000477