Artificial Intelligence for Infrastructure Health Monitoring in Low-Maintenance Environments: A Theoretical Account of Sensing, Inference and Intervention for Roads, Bridges and Power Systems

Abstract

Across much of Africa, infrastructure fails not because its deterioration cannot be detected but because it is not monitored, not maintained, and replaced only after it has collapsed. Artificial intelligence (AI) offers new means of monitoring: low-cost sensors, drone and satellite imagery, smartphone data and machine-learning inference. This paper examines what AI-enabled monitoring can contribute where the maintenance system it is intended to inform is weak. Drawing on structural health monitoring and reliability engineering, on the predictive-maintenance and asset-management literatures, and on recent research on computer vision, sensor networks, satellite interferometry and machine learning for roads, bridges and power systems, it distinguishes three stages in the monitoring-to-maintenance process: sensing, which acquires condition data; inference, which converts those data into a diagnosis and prognosis; and intervention, which acts upon the diagnosis. AI has substantially reduced the cost of sensing and improved inference, but it has left intervention, which is institutional and financial in nature, largely unchanged. The paper therefore argues that the engineering value of monitoring depends on the intervention system it serves, and that in low-maintenance environments monitoring should be designed for the limited set of decisions such systems still make, emergency response, the prioritisation of scarce maintenance budgets, and advocacy for maintenance funding, rather than for comprehensive condition assessment. The data, power, connectivity, security and skills constraints on deployment in Nigerian and African infrastructure are examined, and implications are drawn for infrastructure agencies, utilities, funders and researchers. The paper concludes that, where the capacity to act on condition data is weak, investment in monitoring should follow rather than precede investment in the capacity to intervene, because detection that cannot be acted upon yields information without any improvement in reliability or safety.