Bio-Inspired Compressive Sensing based on Auditory Neural Circuits for Real-time Monitoring and Control of Civil Structures using Resource Constrained Sensor Networks.

Recent natural hazard disasters including Hurricane Sandy (2012) and the Tohoku Earthquake (2011) have called public attention to the vulnerability of civil infrastructure systems. To enhance the resiliency of urban communities, arrays of wireless sensors and actuators have been proposed to monitor and control infrastructure systems in order to limit damage, speed emergency response, and make post-disaster decisions more efficiently. While great advances in the use of wireless sensor networks (WSNs) for the purposes of monitoring and control of civil infrastructure have been made, significant technological barriers have hindered their ability to be reliably used in the field for long durations. Some of these limitations include: reliance on finite, portable power supplies, limited radio bandwidth for data communication, and limited computational capacity. To resolve current bottlenecks, paradigm-altering approaches to the design of wireless monitoring and control systems are required. Through the process of evolution, biological central nervous systems (CNS) have evolved into highly adaptive and robust systems whose sensing and actuation capabilities far surpass the current capabilities of engineered (i.e., man-made) monitoring and control systems. In this dissertation, the mechanisms employed by biological sensory systems serve as sources of inspiration for overcoming the current challenges faced by wireless nodes for structural monitoring and control. The basic, yet elegant, methods of signal processing and data transmission used by the CNS are mimicked in this thesis to enable highly compressed communication with real-time data processing for WSNs engaged in infrastructure monitoring. Specifically, the parallelized time-frequency decomposition of the mammalian cochlea is studied, modeled, and recreated in an ultra-low power analog circuit. In lieu of transmitting data, the cochlea-inspired wireless sensors emulate the neurons by encoding filtered outputs into binary electrical spike trains for highly efficient wireless transmission. These transmitted spike train signals are processed for pattern classification of sensor data to identify structural damage and to perform feedback control in real-time. A key contribution of this thesis is the development and experimental validation of a bio-inspired wireless sensor node that exhibits large energy savings while employing real-time processing techniques, thus overcoming many of the current challenges of traditional wireless sensor nodes.PhDCivil EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/107302/1/cpeckens_1.pd