A Deep Learning Approach for Impact Diagnosis

Abstract

Impact on a structure generates an elastic wave that propagates through the structure carrying a wealth of information about the impact event. The paper presents a deep learning (DL) approach for analyzing these wavefields for the purpose of impact diagnosis i.e., identifying the impact location and reconstructing the impact force time-history. Unlike traditional object detection in computer vision, the nature of the impact diagnosis problem requires capturing context from the wavefield evolution i.e., it involves learning across multiple time frames of the wavefield rather than focusing only on a single stationary frame at a given moment. While scanning across multiple time frames provides essential information about the wave propagation phenomenon in terms of its interactions with structural features, boundaries etc., it mandates the use of deep learning models that can analyze this complex wave propagation phenomenon in both spatial and temporal regimes. A unified CNN-RNN network architecture proposed in the paper first reveals the impact location and then utilizes that information to reconstruct the impact force timehistory. The proposed approach is verified using simulated wavefields obtained from the finite element analysis of a five-bay stiffened aluminum panel. Even with moderate network complexity, the proposed model predicts the impact location and impact force time-history with reasonable accuracy. The potential extension of the proposed methodology to an end-to-end vision-based impact monitoring system is also discussed towards the end of the paper.