Data-Driven Structural Health Monitoring Using Feature Fusion and Hybrid Deep Learning

Smart structural health monitoring (SHM) for large-scale infrastructure is an intriguing subject engineering communities thanks to its significant advantages such as timely damage detection, optimal maintenance strategy, and reduced resource requirement. Yet, it a challenging topic requires handling large amount of collected sensors data continuously, which inevitably contaminated by random noises. Therefore, this study developed practical end-to-end framework that makes use physical features embedded in raw elaborated hybrid deep learning model, namely 1-DCNN-LSTM, featuring two algorithms—convolutional neural network (CNN) long-short term memory (LSTM). In order extract relevant from sensory data, the method combines various signal processing techniques autoregressive discrete wavelet transform, empirical mode decomposition. The 1-DCNN-LSTM designed based on CNN’s capacity capturing local information LSTM network’s prominent ability learn long-term dependencies. Through three case studies involving both experimental synthetic sets, demonstrated proposed approach achieves highly accurate powerful 2-D CNN, but with lower time complexity, making suitable real-time SHM. Note Practitioners —This article aims develop data-driven automatically operational state structures. achieve consistently results performing different tasks diverse structures, we combine underlying frequency domains extracted measured vibration data. Three popular methods are combined diversity gain would not be possible each individual method. As usually long time-series signals, most efficient architecture signal, (LSTM), considered work. Besides, structure has own dynamic properties, i.e., eigenfrequencies, around domain, thus convolutional specifically used combination LSTM, forming architecture. applicability effectiveness supported types showing detection requirements. These can valuable developing model live future life-line infrastructures.