On the Use of Spectral Data from Smartphone Accelerometer Signals and Constituent Material for the Identification of Damaged Walls

Monitoring the integrity of civil structures is crucial for ensuring their safety and longevity. Assessing the structural condition of walls is particularly important due to their role in stability and load distribution in buildings. Smartphones can be used to collect dynamic data from the walls of modern buildings. This strategy is an easier and cheaper way to obtain information concerning the wall's structural condition compared to other costly instrumentation plans using deflectomers, accelerometers, etc. Such a type of data was explored in the literature with good results. However, despite the quality of the models obtained, other features can be included to improve the results. For instance, spectral information may characterize the frequency content of a signal. Moreover, the material used to build the structure affects the signals collected. Thus, we propose the use of three machine learning models (Decision Tree, Random Forest, and K-Nearest Neighbors (KNN)) to identify damage in walls from vibration signals using their spectral data and the wall's constituent material in addition to those already used in the literature. The proposed improvements increased accuracy by about 23%, leading to an average accuracy of 97.78% with KNN when combining statistical and spectral features.