Vibration-Controlled UAV Tap Testing for Predictive Maintenance Using Machine Learning

Abstract

Early detection of mechanical deterioration in critical infrastructure, such as tall buildings with exterior facades, is essential to optimize predictive maintenance strate- gies, reduce operating costs and minimize structural risks. However, conventional inspection methods have significant limitations, including low robustness to environmen- tal disturbances and a high degree of subjectivity that compromises the accuracy of the results. This study proposes an automated tap testing system for acoustic data collection and analysis, specifically designed to simulate the vibrational conditions associated with UAV flight during actual inspections. The approach combines controlled tap test- ing with acoustic detection, acquisition of acoustic signals on specimens with controlled states (healthy/unhealthy) and systematic introduction of vibrations in three amplitudes (1°, 3°, 5°) to evaluate the robustness of the system. Data processing employs Principal Component Analysis (PCA) for dimensional reduction and discriminative feature ex- traction, followed by clustering techniques, k-means, for automatic state classification. Experimental results demonstrate an effective defect discrimination capability, although a progressive degradation of performance proportional to vibration intensity is observed. This work provides a methodological framework for nondestructive inspections under dynamic conditions, laying the groundwork for future improvements in model general- ization and adaptability to complex environments.