Development of a Low-Cost IoT-Based Structural Health Monitoring System for the Maputo-Katembe Suspension Bridge

{ "background": "The Maputo-Katembe suspension bridge is a critical infrastructure link requiring continuous assessment. Traditional structural health monitoring (SHM) systems are often prohibitively expensive for widespread deployment in resource-constrained contexts, creating a need for affordable, reliable alternatives. ", "purpose and objectives": "This short report details the development and pilot deployment of a low-cost, Internet of Things (IoT) -based SHM system. The primary objective was to create a system capable of real-time vibration monitoring and anomaly detection using accessible components. ", "methodology": "A network of low-cost, MEMS-based accelerometers was installed on the bridge's main cables and deck. Data was transmitted via a long-range wide-area network to a cloud platform. A machine learning pipeline was implemented, featuring a principal component analysis (PCA) model for dimensionality reduction, defined as =, where is the standardised sensor data matrix and contains the eigenvectors of the covariance matrix. An isolation forest algorithm was then trained on the principal components to identify anomalous vibrational patterns. ", "findings": "The system successfully collected continuous acceleration data over an extended monitoring period. The isolation forest model identified anomalous events with a calculated precision of 0. 87 (95% CI: 0. 82, 0. 91). A predominant theme from the data was the clear correlation between heavy goods vehicle traffic and specific higher-frequency deck vibrations. ", "conclusion": "The developed system demonstrates the feasibility of using low-cost IoT sensors for effective real-time SHM on major suspension bridges. The integration of PCA and machine learning provides a robust method for automated anomaly detection from high-dimensional sensor data. ", "recommendations": "Future work should focus on long-term reliability testing of the sensor nodes and expanding the machine learning model to incorporate environmental data (e. g. , temperature, wind) to reduce false positives. The methodology is recommended for consideration on similar infrastructure projects across the region. ", "key words": "structural