Advancement of bridge health monitoring using magnetostrictive sensor with machine learning techniques

With the increasing demand for sustainable infrastructure maintenance, it has become very important to predict the health condition of the structures in real-time. This study investigates the application of machine learning techniques for assessing the structural health of prototype beam bridges. By employing magnetostrictive sensors, which convert mechanical vibrations into electrical energy, the research aims to perform frequency analysis to predict dominant frequencies in a prototype beam bridge. Data were collected using a Digital Storage Oscilloscope and a Data Acquisition Card, followed by comprehensive feature extraction and dimensionality reduction. Machine learning models, including Random Forest and Deep Neural Networks, were utilised to classify waveform types and predict vibration frequencies. The Random Forest model achieved a classification accuracy of 86.16% and a mean absolute percentage error of 4.33% in frequency prediction, highlighting its superior accuracy and reliability for continuous bridge health monitoring. These results demonstrate the potential to revolutionise modern infrastructure maintenance practices by enabling real-time, automated assessments of structural integrity.