Structural health monitoring on offshore jacket platforms using a novel ensemble deep learning model

Monitoring health condition of offshore jacket platforms is crucial to prevent unexpected structural damages, where a prevailing challenge involves translating available feature information into structural damage patterns. Although the artificial neural network (ANN) models are popular in addressing this challenge, they often fail to capture the temporal correlations between the feature information and the damage patterns, which reduce their capability for discovering the laws governing the structural damage detection. To bridge this research gap, this study proposes a novel ensemble deep learning model to enhance the temporal feature extraction to improve the damage pattern identification. In this approach, a one-dimensional Convolutional Neural Network (CNN) extracts the spatiotemporal features from the structural vibration measurements. Simultaneously, a SENet attention mechanism is introduced to select the most informatic features. Subsequently, a bidirectional long short-term memory network (BiLSTM) is employed to learn the mapping between the extracted features and the structural damage patterns. Furthermore, the particle swarm optimization (PSO) algorithm is used to optimize the BiLSTM hyperparameters to enhance its stability and reliability. Both simulations and experiments are carried out to collect the vibration responses of the offshore jacket structure in different damage scenarios. The analysis results demonstrate that the proposed method produces remarkable improvement with respect to the accuracy and robustness in identifying the structural damages when compared with the ANNs. The overall detection accuracy of the proposed CNN-BiLSTM-Attention ensemble model is beyond 95%, which provides strong applicability to practical structural health monitoring of offshore platforms.