Real-time hysteresis identification in controlled structures based on restoring force reconstruction and Kalman filter

尽管在识别结构中的滞回力方面取得了很大进展，但由于逆问题的病态性，仍然可能识别出具有较大误差的非线性分量重建中不令人满意的滞回行为。今年，我们改进了基于卡尔曼滤波的恢复力重建方法，将深度神经网络引入到经典的数值积分方法中，采用混合积分时间步进方法。我们提出用残差网络对系统的非线性行为进行一般辨识。与之前的方法相比，新开发的Physics-DNN杂交积分时间步进方案在迟滞识别和相应的结构动力分析方面都提供了稳定的解决方案。

Despite great progress in identifying the hysteretic forces in structures, less satisfactory hysteretic behaviors with large errors for the nonlinear component reconstruction might still be identified, due to the ill-posedness of the inverse problem. This year, we improved the Kalman filter based restoring force reconstruction method through incorporating deep neural networks into the classical numerical integration method by using a hybridized integration time-stepper. We proposed to use residual network to generally identify the nonlinear behaviors of the system. Compared to the previous method, the newly developed Physics-DNN hybridized integration time-stepping scheme provides stable solution in both of the hysteresis identification and the corresponding structural dynamic analysis.