A Systematic Literature Review on the Engineering Architecture of AI-Driven Systems for Predicting Infrastructure Material Degradation and Remaining Useful Life

Abstract

Infrastructure aging poses severe structural integrity risks, yet traditional monitoring often relies on low-frequency data acquisition that fails to capture complex, non-linear material degradation. This study systematically reviews the engineering deployment of Artificial Intelligence (AI) models within Structural Health Monitoring (SHM) and Asset Information Systems to predict material degradation and Remaining Useful Life (RUL). Using the PRISMA methodology, we analyzed peer-reviewed articles focusing on system architecture and algorithmic performance. The findings indicate a paradigm shift toward hybrid Deep Learning models (e.g., CNN-LSTM) to process high-frequency spatial-temporal sensor data. However, critical engineering bottlenecks remain, particularly regarding real-time sensor telemetry, MLOps integration in legacy systems, and edge-computing constraints. In conclusion, transitioning predictive models from isolated laboratory environments to robust, scalable engineering systems is imperative for ensuring the physical integrity and reliability of critical infrastructure.