Optimization of Reinforced Concrete Bridge Deck Health Monitoring Using SHM Techniques and Machine Learning

Project Abstract

This study presents a comprehensive framework for health monitoring of reinforced concrete bridge decks by integrating structural health monitoring (SHM) techniques with data-driven machine learning models to enhance safety, performance, and maintenance planning. The research addresses the critical need for real-time assessment of deterioration processes such as cracking, corrosion, delamination, and moisture ingress, which traditionally rely on periodic visual inspections and nondestructive testing methods with limited temporal resolution. A hybrid sensing network is developed, combining embedded piezoelectric transducers for active impact-echo and guided wave testing, surface-mounted accelerometers for vibration-based monitoring, humidity and temperature sensors, and wireless data loggers to enable continuous data acquisition under varying environmental and loading conditions. The experimental phase includes calibration on laboratory concrete slabs with controlled defect populations, followed by deployment on representative bridge decks subjected to realistic loading schemes and environmental exposure. Signal processing and feature extraction are performed to capture damage-sensitive indicators, including wave dispersion characteristics, impedance-based metrics, modal parameters, crack propagation proxies, and moisture-related impedance changes. These features feed into several machine learning architectures, such as supervised deep neural networks for damage classification and regression, ensemble methods for robust performance under data imbalance, and semi-supervised learning to leverage limited labeled data. A novel transfer learning strategy is introduced to adapt models trained on one bridge deck to another with similar structural typologies, thereby reducing the data collection burden for new projects. The methodology emphasizes model explainability and uncertainty quantification through probabilistic regression and SHAP-based feature importance analysis to ensure decision-makers can interpret predictions with confidence. The optimization component focuses on developing a decision-support framework that prioritizes inspection and maintenance actions according to risk, cost, and downtime constraints, using multi-objective optimization and Bayesian updating as new data arrive. The study also investigates the fusion of SHM-derived indicators with traditional structural performance metrics to improve early warning capabilities for critical defects. Results demonstrate improved detection sensitivity for subcritical damage, earlier initiation of corrective actions, and a reduction in unnecessary maintenance interventions compared to conventional inspection programs. Sensitivity analyses reveal how sensor placement, data quality, and environmental variability influence model performance, guiding practical deployment strategies. The proposed framework is validated against field data from multi-year bridge monitoring campaigns, with cross-validation across different seasons and loading scenarios. Findings indicate that integrating SHM with machine learning enables continuous, scalable, and context-aware health assessment, ultimately contributing to safer bridge operation, extended service life, and optimized lifecycle costs. The research concludes with practical guidelines for implementation, sensor network design, data management, model maintenance, and policy implications for infrastructure resilience in the face of aging bridge stock and climate-induced deterioration.

Project Overview

What This Project Is About

---

A plain-language overview of monitoring the health of reinforced concrete bridge decks using sensor-based data. The project combines structural monitoring methods (how the bridge behaves under loads and over time) with data analysis techniques to detect early signs of damage and degradation.

---

---

The Problem It Addresses

---

Bridge decks can develop cracks, corrosion, and stiffness loss that are hard to notice until they worsen. Traditional inspections are periodic and may miss rapid changes. This project aims to provide continuous, real-time insight into deck health to improve safety and maintenance decisions.

---

---

Objectives of the Project

---

1. Explain the basics of SHM and how sensors can track deck performance.
2. Develop a simple data collection plan using affordable sensors.
3. Implement machine learning to interpret sensor data for damage indicators.
4. Show how the approach can predict remaining service life or maintenance needs.
5. Evaluate the method on a case study or simulated deck data.

---

---

What You Will Do Step by Step

---

1. Learn key SHM concepts and select suitable sensors.

2. Design a data collection protocol for deck monitoring.

3. Collect or simulate data, including normal and damaged conditions.

4. Preprocess data and extract meaningful features (e.g., vibration, deflection).

5. Train a simple machine learning model to classify health states.

6. Validate the model with test data and assess accuracy.

7. Discuss how results support maintenance decisions.

8. Reflect on limitations and future improvements.

---

---

Expected Outcome

A practical framework that uses SHM data and a lightweight machine learning model to identify bridge deck health issues early, with clear recommendations for maintenance timing and further study.