A fast-response-generation method for single-layer reticulated shells based on implicit parameter model of generative adversarial networks

Single-layer reticulated shells (SLRSs) are used in the construction of the roofs of important public buildings, such as gymnasiums and exhibition center; thus, it is crucial to monitor, evaluate, and predict their health status regularly. However, the response data obtained using the monitoring system are often inadequate. Therefore, the finite element (FE) model is modified according to the existing monitoring data, and the structural response is obtained through FE analysis. However, the process is time-consuming, and the calculated response is not necessarily accurate since the FE model includes many assumptions and simplifications. In this paper, a fast-response-generation method for SLRSs based on an implicit parameter model of generative adversarial networks (GANs) is proposed. First, based on the time-domain response data obtained from monitoring, frequency-domain response data including amplitude, phase, and frequency are calculated by Fourier transform, and the implicit parameter model based on frequency-domain response of SLRS is proposed. The frequency-domain response data are then transformed into the GAN dataset by the dataset construction method based on frequency-domain response. On the basis of this constructed dataset, the corresponding GAN architecture, namely, the implicit parameter model, is built. After the implicit parameter model is trained, the frequency-domain response of the structure can be obtained quickly, followed by the time-domain response through inverse Fourier transform. Finally, the validity of the proposed implicit parameter model is verified by a numerical spherical SLRS and an actual SLRS. The numerical example and application example show that the GAN can generate realistic frequency-domain response, and the generated responses are similar and diverse compared with actual responses. By using certain vibration data obtained from the structural health monitoring system as the training dataset, the constructed implicit parameter model can replace the FE model to generate vibration responses rapidly. •The implicit parameter model of generative adversarial networks (GANs) for the single-layer reticulated shells (SLRSs).•The GAN’s dataset construction method based on frequency-domain response.•GANs for different DOFs that can quickly generate frequency-domain response similar to corresponding actual response.•A fast-response-generation method for SLRSs based on implicit parameter model of GANs.