A Novel Machine Learning Framework for Real-Time Structural Health Monitoring and Damage Classification of Nigeria’s Critical Highway Bridge Network

Nigeria's critical highway bridge network is monitored through periodic manual inspection at intervals that are too infrequent to capture the onset and progression of damage between inspection cycles, creating structural safety risks for bridges carrying strategic economic and humanitarian traffic. This study develops a novel machine learning framework for real-time structural health monitoring and automated damage classification applicable to the Nigerian highway bridge network, addressing the twin challenges of sparse sensor network operation under power-constrained field conditions and the need for interpretable classification outputs that inform maintenance decision-making by engineers without machine learning expertise. The framework integrates a physics-informed autoencoder for anomaly detection in vibration data, a convolutional neural network for damage typology classification trained on a synthetic training dataset generated from validated finite element bridge models, and a Bayesian model updating module for damage localisation under uncertainty. The framework was trained on vibration data generated from finite element models of three representative Nigerian bridge typologies and validated on field data from three instrumented bridges in Ogun, Kano, and Benue States representing South West, North West, and North Central geopolitical zones. Novel contributions include the development of a transfer learning approach for adapting the classification model to new bridge structures using fewer than 20 labelled field observations, enabling rapid deployment across the network without extensive re-training. Damage classification accuracy on the validation bridge dataset achieved 89 percent at the damage type level and 76 percent at the damage location level. The study provides a deployable framework and open-source software implementation for progressive rollout under the Federal Ministry of Works bridge monitoring programme. Keywords: machine learning, structural health monitoring, bridge damage classification, highway bridges, transfer learning.