TLDR: This study introduces an integrated system combining a LoRa-based Internet of Things (IoT) network with machine learning to predict corrosion rates in steel heritage structures, like the San Sebastian Basilica. By monitoring only temperature and humidity, the system overcomes limitations of traditional methods that require extensive pollutant data. It uses ensemble regression models (Random Forest, Gradient Boosting, XGBoost) to accurately forecast corrosion, with Random Forest showing the highest precision. The framework is deployed via a real-time dashboard, offering a scalable and cost-effective solution for proactive preservation worldwide.
Cultural heritage structures, such as the magnificent San Sebastian Basilica in the Philippines, are invaluable testaments to our history and identity. However, these steel-built landmarks face a constant threat: atmospheric corrosion. Traditional methods for assessing corrosion risk often rely on extensive data about air pollutants, which can be difficult and costly to collect consistently at heritage sites.
A recent study, titled Corrosion Risk Estimation for Heritage Preservation: An Internet of Things and Machine Learning Approach Using Temperature and Humidity, by Reginald Juan M. Mercado, Muhammad Kabeer, Haider Al-Obaidy, and Rosdiadee Nordin, introduces an innovative solution to this challenge. Their research proposes an integrated framework that combines Internet of Things (IoT) technology with advanced machine learning to provide accurate and proactive corrosion forecasting using minimal environmental data.
A Smart Monitoring System for Heritage Sites
The core of this new approach is a custom-designed, LoRa-based IoT hardware system. LoRa (Long Range) wireless communication is ideal for heritage sites because it allows for long-range, low-power monitoring of environmental parameters without needing to alter the structure itself. For the San Sebastian Basilica, a network of 14 sensor stations and one LoRa gateway was deployed. These sensors continuously monitor crucial environmental factors like temperature and relative humidity, which are primary drivers of corrosion.
This system collects a robust three-year dataset, providing a rich foundation for understanding how environmental conditions affect steel structures over time. The data collection is synchronized, ensuring high-resolution and reliable information that links indoor microclimates with external weather conditions.
Predicting Corrosion with Machine Learning
With the environmental data in hand, the researchers developed a sophisticated machine learning framework. This framework is designed to predict atmospheric corrosion rates using only temperature and relative humidity measurements, overcoming the limitations of traditional methods that require pollutant concentration data. Three ensemble regression models were employed: Random Forest, Gradient Boosting, and XGBoost.
These models were trained on the extensive dataset, using rigorous time-series validation to ensure their predictions would be realistic in real-world forecasting scenarios. All three models showed exceptional predictive performance. Random Forest emerged as the most accurate, achieving an R² value of 0.9913, indicating it could explain over 99% of the variance in corrosion rates. Gradient Boosting and XGBoost also performed strongly, with R² values of 0.9902 and 0.9881, respectively.
While Random Forest offered the highest precision, XGBoost demonstrated superior computational efficiency, training in just 2.62 seconds compared to Random Forest’s 102.92 seconds. This balance of accuracy and speed means that different models can be chosen based on the specific needs of a conservation project.
Real-time Insights for Proactive Preservation
To make these predictions actionable, the framework is deployed via a Streamlit dashboard, made publicly accessible through ngrok tunneling. This dashboard provides real-time corrosion monitoring and translates predictions into actionable risk classifications. For instance, it can issue alerts like “activate dehumidifiers” when conditions suggest an increased corrosion risk, enabling maintenance teams to take preventive action before significant damage occurs.
This minimal-data approach offers a scalable and cost-effective solution, particularly beneficial for heritage sites with limited monitoring resources. It demonstrates that advanced regression techniques can extract highly accurate corrosion predictions from basic meteorological data, empowering proactive preservation strategies for culturally significant structures worldwide without requiring extensive sensor networks.
Also Read:
- AI Agents Accelerate New Alloy Development for 3D Printing
- Enhancing Maritime Shaft Power Prediction with Cross-Frequency Transfer Learning
Looking Ahead
This study successfully bridges the gap between theoretical corrosion science and practical heritage preservation. The researchers plan to further enhance the system by developing robust data preprocessing pipelines to handle missing sensor data and expand validation across diverse heritage sites. They also aim to explore integrating additional sensor modalities, such as airborne pollutant concentrations, to further refine the models while maintaining cost-effectiveness.
