Dynamic data reconstruction of Structural Health Monitoring (SHM) using VMD And Depthwise-1D-CNN

Abstract Effective structural health monitoring systems rely on accurately acquired data, yet data loss is inevitable, often stemming from various factors such as system failures or power outages across the four monitoring stages: data collection, communication, storage, and evaluation. To address this challenge, this paper presents a novel approach for dynamic response data recovery employing variational mode decomposition (VMD) combined with depth-wise 1-dimensional Convolutional Neural Network (Depthwise-1D-CNN). Specifically, VMD decomposes time-series data into more stable signals known as intrinsic mode functions (IMFs), followed by the application of Depthwise-1D-CNN for prediction. The final outcome aggregates all predicted IMFs. The effectiveness of the proposed method is demonstrated through a case study utilizing a numerical model employing an I-section steel beam. Results indicate a significant improvement in terms of root mean square error and mean absolute error by around 85%.