Locating Impacts Through Structural Vibrations Using the FEEL Algorithm Without a Known Input Force

Abstract

Floor vibration-based methods to track human activity are becoming popular for applications in healthcare monitoring, security, and occupant detection. Popular techniques such as time of arrival (TOA) methods face wave dispersion and multiple-path fading challenges for localization. Data-driven methodologies such as the FEEL Algorithm rely exclusively on the system dynamic properties, an advantage over other methods. However, FEEL’s calibration process requires recording force input to the structure, which can become labor-intensive and time-consuming for applications that require a high localization accuracy and does not require force estimates. An alternative approach is proposed to use the system’s acceleration response exclusively, creating an output-to-output transfer function. This modification was tested against the 3575 impact Human-Induced Vibration Benchmark dataset containing seven impact types across five locations, the same dataset FEEL was originally developed with. The results demonstrated the acceleration-calibrated FEEL effectiveness with 99.9% localization accuracy compared to force-calibrated FEEL’s accuracy of 96.4%.