Estimation of Structural Vibration Modal Properties Using a Spike-Based Computing Paradigm

Spiking neural networks are an emerging concept that draws inspiration from the computational neuroscience research community. Spiking neural networks combine spike-based computing and machine-learning-based neural networks that emulate the operation of the human brain. Spiking neural networks have the ability to be easily integrated into neuromorphic hardware, such as Intel’s Loihi chip. The advantages of neuromorphic hardware are its high-speed computation and low-power consumption in comparison to traditional electronics. These factors have an important role for the future of smart systems and the reliability of structural health monitoring. Currently, coupling spike-based computing to continuous-valued signals, which are typically measured in structural dynamics, is rare. This paper aims to explore spiking neural networks and their possible application in structural dynamics and modal analysis using Nengo, a large-scale neural network simulation package. In this work, we implement output-only modal identification techniques that rely on solving the blind source separation problem using spike neural networks to extract the natural frequencies, mode shapes, and damping ratios of a simulated structural system being exposed to dynamic loading.