Search

The deployment of systems experiencing high-rate dynamic events, such as hypersonic vehicles, advanced weaponry, and active blast mitigation systems, require high-rate structural health monitoring (HRSHM) capabilities in the sub-millisecond realm to ensure continuous operations and safety. However, the development of high-rate feedback systems is a complex task because these dynamic systems are uniquely characterized by (1) large uncertainties in their external loads, (2) high levels of non-stationarity and heavy disturbance, and (3) unmodeled dynamics from changes in system configuration. In this paper, we present a deep learning algorithm specifically engineered for HRSHM applications. It consists of an ensemble of recurrent neural networks (RNNs) constructed with long short-term memory (LSTM) cells with transfer learning capabilities to cope with the highly limited availability of training data as it is typical for high-rate systems. The use of an ensemble of RNNs empowers multi-rate sampling capability to capture multi-temporal features of the time series, thus enabling modeling of non-stationarities. Also, because the RNNs use short-sequence LSTMs and are arranged in parallel, the computation time is substantially reduced to the sub-millisecond range. The performance of the algorithm is investigated on experimental high-rate dynamic data produced from accelerated drop tower tests and benchmarked against results from a prior investigation using a purely on-the-edge algorithm termed variable input space observer (VIO) and its hybrid architecture (hybrid VIO) that incorporated physical knowledge and is considered as an upper bound on the algorithm performance. Numerical results show that the ensemble of RNNs, composed of five RNNs, significantly outperformed the VIO, with performance close to that of the hybrid VIO when it comes to the estimation error metrics, with an average computation time of 25 μ s per prediction step. Yet, the ensemble of RNNs exhibits chattering for low-amplitude excitations, likely attributable to the behavior of RNNs sampling at small time delays. An examination of the RNN weights and hidden states confirms that the algorithm can capture multi-temporal features, and an investigation with respect to noise in the training dataset showed that the algorithm is robust for signal-to-noise ratios up to 20 dB. • Machine learning-based architecture for non-stationary time-series analysis. • Transfer learning for accelerated training with limited data. • Ensemble-based system estimation using parallel recurrent neural networks integrated with attention mechanism is proposed. • Sib-millisecond estimation of a high-rate system is achieved though the proposed architecture. [ABSTRACT FROM AUTHOR]

This paper proposes a novel technique that aims at detecting the effect of damage on structural frequency signals as "bad" outliers. To this end, a procedure is developed based on the Variational Mode Decomposition (VMD), Minimum Covariance Determinant (MCD), and Recurrent Neural Network (RNN) with Bi-directional Long-Short Term Memory (BiLSTM) cells. The VMD is first used in a pre-processing stage to denoise the signals and remove the seasonal patterns in them. Then, the proposed method seeks to learn the rules behind calculation of the Mahalanobis distances of the points from their distribution, using the parameters obtained from the MCD algorithm, through training an RNN on signals obtained from the inferior state of the structure (healthy state). It will be shown that, since the rule behind the effect of damage on the Mahalanobis distances has not been learnt by the trained RNN, the prediction errors of these values will increase significantly as soon as damage occurs using the data obtained from the posterior state of the structure (including damage). The performance of the proposed method is first tested on a numerical example and further validated through solving an experimental example of the Z24 bridge. Moreover, the proposed method is compared against a PCA-based method. The results demonstrate the superiority of the proposed method in long-term condition monitoring of civil infrastructures. The proposed method is an output-only condition monitoring method that requires only a couple of lowest structural natural frequency signals measured over a long-term monitoring of the structure. Therefore, it is recommended for cases when the measurements from the EOV are not available. Also the proposed method can be used along with other output-only or input-out methods to either improve or confirm the validity of their results. • A new condition monitoring method of the structures under EOV is proposed. • VMD is used to denoise the frequency signals while removing the seasonal pattern. • An RNN is trained to learn the rule behind calculation of the Mahalanobis distances. • The method is tested on both numerical and experimental examples. • The prediction error of Mahalanobis distance increases as damage occurs. [ABSTRACT FROM AUTHOR]