Comparing Statistical Classification with a Vibro-Tactile Human–Machine Interface for Structural Health Monitoring

Abstract

A goal of the structural health monitoring (SHM) community has been to endow structures with a nervous system not unlike that present in biologic organisms. Typically in SHM, data from sensors is collected then analyzed using statistical classifiers and/or compared to results from physics based models. Despite successes using these techniques, they are not, in general, adaptable to unforeseen loading and/or ambient conditions. Past studies have shown that, in some cases, humans can outperform computers in classification under such conditions. In this work, we are interested in testing an interface between SHM data and humans using a vibro-tactile device. Research from the neuroscience community suggests this interface may be possible by leveraging the phenomena of sensory substitution. To test the hypothesis, an excitation was provided to a three-story structure, and the output acceleration response for each floor was collected. The structure could be modified with bumpers to exhibit non-linear dynamic responses on any combination of its three floors. The human subject wore a vibro-tactile glove excited by the acceleration data and reported which bumpers they believed were engaged based upon the stimulation of the actuators. The performance of human subjects to identify the cause of non-linearity in structures was compared with results achieved using a support vector machine classifier.