Structural damage inverse detection from noisy vibration measurement with physics-informed neural networks

Abstract

Structural damage detection is an inverse problem to identify and quantify structural damage from measurement data by discovering the variation of structural mechanical parameters. Recently, a novel deep learning framework named physics-informed neural networks (PINNs) has been proposed and successfully applied to solve inverse problems of various linear/nonlinear partial differential equations (PDEs) by integrating physical information such as governing equations as prior information. In this study, we propose a PINN-based framework to exploit a novel method of structural damage detection. Specifically, a deep neural network model as the core of PINNs is built to predict the dynamic response in different degrees of freedom. The unknown mechanical parameters are initialized randomly and updated together with the neural network model parameters. Then, the structural physics model is embedded by calculating the residuals of governing equations as parts of the loss function. The residual between the predicted dynamic response and measurement data is also used as another part of the loss function. A two-step optimization strategy is proposed to obtain the best unknown parameter values that can fit the measurement data and governing equations simultaneously. Through numerical experiments of a single-degree-of-freedom system, we demonstrate that the proposed method can successfully identify potential structural mechanical parameters and quantitatively detect structural damage. The influence of sparsity and noise in the measurement data on the detection results is also analysed.