The purpose of this study is to discuss the possibility of the concept of physical reservoir computing (PRC) in the field of structural health monitoring (SHM). PRC is a physical realization of a class of recurrent neural networks called reservoir computing (RC). This consists of an input layer, mutually connected network of neurons with strong nonlinearity with fixed coupling weights (referred to as reservoir), and an output layer with learnable weights. The key idea of PRC is to replace the reservoir part in RC by a specific physical entity, which has opened new possibilities of smart structures by providing a way to embed some sort of intelligence in structures. In this study, we propose to apply this framework to SHM by regarding the target structure itself as the physical reservoir. Unlike the conventional problem setting in PRC, our purpose is to detect the change occurred in the physical reservoir due to structural failure. In this paper, we propose one possible methodology to achieve this, in which the output layer is trained to learn some nonlinear function so that the increase of the error may indicate the change of the reservoir due to failure. A simple toy problem using a network of interconnected nonlinear oscillators are presented to examine the validity of the proposed method.

© (2023) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.

Arata Masuda, Konosuke Takashima, and Ryu Sakai, "Physical reservoir-based structural health monitoring: a preliminary study," Proc. SPIE 12483, Active and Passive Smart Structures and Integrated Systems XVII, 1248311 (Presented at SPIE Smart Structures + Nondestructive Evaluation: March 16, 2023; Published: 28 April 2023); https://doi.org/10.1117/12.2658742.