The standard camera- and force plate-based set-up for motion analysis suffers from the disadvantage of being limited to laboratory settings. Since adaptive algorithms are able to learn the connection between known inputs and outputs and generalise this knowledge to unknown data, these algorithms can be used to leverage motion analysis outside the laboratory. In most biomechanical applications, feedforward neural networks are used, although these networks can only work on time normalised data, while recurrent neural networks can be used for real time applications. Therefore, this study compares the performance of these two kinds of neural networks on the prediction of ground reaction force and joint moments of the lower limbs during gait based on joint angles determined by optical motion capture as input data. The accuracy of both networks when generalising to new data was assessed using the normalised root-mean-squared error, the root-mean-squared error and the correlation coefficient as evaluation metrics. Both neural networks demonstrated a high performance and good capabilities to generalise to new data. The mean prediction accuracy over all parameters applying a feedforward network was higher (r = 0.963) than using a recurrent long short-term memory network (r = 0.935).

Summary: Earthquake dynamic response analysis of large complex structures, especially in the presence of nonlinearities, usually turns out to be computationally expensive. In this paper, the methodical developments of a new model order reduction strategy (MOR) based on the proper orthogonal decomposition (POD) method as well as its practical applicability to a realistic building structure are presented. The seismic performance of the building structure, a medical complex, is to be improved by means of base isolation realized by frictional pendulum bearings. According to the new introduced MOR strategy, a set of deterministic POD modes (transformation matrix) is assembled, which is derived based on the information of parts of the response history, so‐called snapshots, of the structure under a representative earthquake excitation. Subsequently, this transformation matrix is utilized to create reduced‐order models of the structure subjected to different earthquake excitations. These sets of nonlinear low‐order representations are now solved in a fractional amount of time in comparison with the computations of the full (non‐reduced) systems. The results demonstrate accurate approximations of the physical (full) responses by means of this new MOR strategy if the probable behavior of the structure has already been captured in the POD snapshots. Copyright © 2016 The Authors. Earthquake Engineering & Structural Dynamics Published by John Wiley & Sons Ltd.

Evaluating the response statistics of nonlinear structures constitutes a key issue in engineering design. Hereby, the Monte Carlo method has proven useful, although the computational cost turns out to be considerably high. In particular, around the design point of the system near structural failure, a reliable estimation of the statistics is unfeasible for complex high‐dimensional systems. Thus, in this paper, we develop a machine‐learning‐enhanced Monte Carlo simulation strategy for nonlinear behaving engineering structures. A neural network learns the response behavior of the structure subjected to an initial nonstationary ground excitation subset, which is generated based on the spectral properties of a chosen ground acceleration record. Then using the superior computational efficiency of the neural network, it is possible to predict the response statistics of the full sample set, which is considerably larger than the initial training sample set. To ensure a reliable neural network response prediction in case of rare events near structural failure, we propose to extend the initial training sample set increasing the variance of the intensity. We show that using this extended initial sample set enables a reliable prediction of the response statistics, even in the tail end of the distribution.HIGHLIGHTS A new Monte Carlo method is developed that provides the response statistics of a nonlinear system in the tail end of the distribution. A nonstationary filter is applied to a Gaussian white noise to generate realistic artificial earthquake records. An extended training strategy using neural networks is proposed to improve the reliability of the method in the tail end of the distribution. The new strategy reveals a significant speedup as well as a prediction of the response statistics in the tail end with high confidence.