Data-driven Pattern Recognition Model Employing Auditory Receptors for Human-based Structural Health Monitoring System

Abstract

A quintessential conceptual element common to most Structural Health Monitoring (SHM) systems is the use of non-destructive methods and technologies to allow for the uninterrupted and efficient monitoring of structural damages. Recent advancements pertaining to the logistical underpinnings of novel damage detection SHM techniques have caused a shift from mathematical modeling of structures to pattern recognition algorithms encompassing both supervised and unsupervised learning methods. This work draws inspiration from current progressions in the field of neuroscience to incorporate the human brain in performing supervised pattern recognition, whereby the initial (i.e. damaged or undamaged) state of the structure is of prior knowledge. In this context, a framework for damage detection relating to shear-building structures has been developed for practical implementation in response to a range of different damage circumstances. The proposed framework uses human ears as auditory receptors to communicate received data to the human brain for interpretation. Two distinct audio signals are derived from the modal analysis of the structure. Induced audible signals are exposed to human subjects, and through consistent training, the ultimate ability of subjects to determine the state of structures is evaluated through training-test schemes for gathered responses. The results obtained are then contrasted with those obtained through the use of specific Machine Learning (ML) algorithms. Initial results involving direct human perception have thus far revealed a noteworthy impact effected by these assessments on understanding the behavior of structures. Possible areas of future growth may also be exploitable if other receptors (for instance, touch receptors as vehicles to transfer data to the human brain through vibrotactile patterns) are used in similar pattern recognition SHM systems