Machine‐learning‐based methods for output‐only structural modal identification

In this study, we propose a machine-learning-based approach to identify the modal parameters of output-only data for structural health monitoring (SHM) that makes full use characteristic independence responses and principle machine learning. By taking advantage independent feature each mode, unsupervised learning, turning training process neural network into separation. A self-coding is designed from vibration structures. The mixture signals, is, response data, are used as input network. Then, complex loss function restrict network, making output third layer want, weights last two layers mode shapes. essentially nonlinear objective optimization problem. novel proposed constrain with consideration uncorrelation non-Gaussianity obtain parameters. numerical example simple structure carried out illustrate parameter identification ability considering influence damping ratios. method further verified by an actual SHM dataset cable-stayed bridge. results show capable blindly extracting information system responses.