Machine Learning-Based Estimation of Modal Properties of a Transmission Tower in Bataan, Philippines

Abstract

Abstract – Although fully automated modal analysis, a Structural Health Monitoring (SHM) technique, has recently been used to monitor the current condition of various civil structures, its application to wind-sensitive transmission towers remains limited. Most modal analysis and dynamic characterization studies related to these towers, which are essential for first-level damage detection, still require manual selection of input parameter values. This paper aims to contribute to the existing discussion by applying a machine learning (ML) algorithm to Stochastic Subspace Identification (SSI) to derive the modal parameters of a transmission tower located in Orion, Bataan, Philippines, thereby enhancing existing methodologies. In addition to utilizing Random Forest as the core intelligence of the method, the research explores three other ML algorithms—XGBoost, Decision Trees, and k-Nearest Neighbors (KNN)—as alternative modal prediction models within the framework. Despite limitations in sensor placement—restricted to the tower’s lower half—the study successfully extracted ten frequency values from an actual transmission tower, closely aligning with analytical predictions. The first mode from the field data was identified at 3.21 Hz, with only a 0.63% deviation from the analytical model. Damping ratios, ranging from 0.68% to 3.02%, exhibited deviations of up to 138% for the fundamental mode but remained within international code recommendations, such as the ASCE 74 guideline of 4%. Random Forest stands out among the ML models tested, showing the fastest runtime, highest performance accuracy, and smallest Coefficient of Variation (CoV) values given random datasets, closely followed by XGBoost. A variability analysis over 120 two-minute datasets showed frequency CoVs between 0.42% and 2.39%, and damping CoVs between 2.02% and 7.02%. The results of this study can be used in model updating and the structural design of transmission towers in the Philippines. They also serve as a baseline for future recordings, facilitating enhanced and data-driven post-disaster decision-making.

Keywords: automated modal analysis, transmission tower, machine learning, structural health monitoring, modal parameters