Multi-LSTM-Based Framework for Ambient Intelligence Nur Sila Gulgec, Martin Takácˇ, and Shamim N. Pakzad Abstract Bridge structures experience significant vibrations, repeated stress cycles, and deterioration during their life cycles. Thus, it is essential to accurately establish life cycle assessment which often needs collecting strain responses. Conventional way of measuring strain measurements is laboriously ineffective and expensive as more spatial information is desired. In the presented approach, inexpensively collected acceleration responses by using wireless sensor networks or mobile sensing are utilized to predict strain information and decrease the installation expenses of sensors systems, as well as enable crowdsourcing potential of the mobile sensing. The study employs a deep learning framework composed of fully connected (FC) and multiple Long Short-Term Memory (LSTM) layers to predict strain time series from the acceleration responses. The deep architecture achieves to learn the relationship between input and output by exploiting the temporal dependencies of them. In the evaluation of the method, acceleration data collected from a steel bridge is utilized to predict the strain time series. The learned architecture is tested on an acceleration time series that the structure has never experienced. The initial findings show that this study holds great potential to perform sensing of the structures with affordable and intelligent sensing systems. Keywords Structural health monitoring · Deep learning · Sensing · Sensor networks · Mobile sensing · Long short-term memory 1 Introduction The life cycle assessment requires designing computationally affordable performance indicators for both the components of the structure and its overall system. Building finite element (FE) models of structures’ systems and components help engineers to provide more accurate performance assessment. Traditionally, structural health monitoring (SHM) data is used to update FE models and lifetime reliability of structures [1]. In reliability analysis, the primary monitored response quantity measured by SHM is strain [2–4]. The life cycle performance is updated by the collected SHM data and the maintenance optimization is performed to assess structures’ life cycle and update its service life. Traditional inspection methods collect strain measurements by using strain; nevertheless, large-scale and spatially dense installation of wired strain gauges is expensive and laboriously impractical [5]. To address these limitations, the integration of information from inexpensive data sources is necessary [6, 7]. Acceleration data can be collected relatively inexpensively by means of fixed acceleration sensors, wireless sensor networks, and mobile sensing. Mobile sensing is an alternative paradigm to collect comprehensive spatial information by using a few sensors. Mobile sensors have low setup costs compared to the traditional stationary sensor networks and they do not require to be dedicated to any particular structure [8]. Most importantly, when the ubiquity of smartphones with the internet of things (IoT) connectivity is considered, cars+smart phones can be treated as large-scale sensor networks which can contribute to the life cycle assessment of the structures on a daily basis [9]. N. S. Gulgec ( ) Thornton Tomasetti, San Francisco, CA, USA M. Takácˇ Department of Industrial and Systems Engineering, Lehigh University, Bethlehem, PA, USA e-mail: mat614@lehigh.edu S. N. Pakzad Department of Civil and Environmental Engineering, Lehigh University, Bethlehem, PA, USA e-mail: snp208@lehigh.edu © The Society for Experimental Mechanics, Inc. 2022 K. Grimmelsman (ed.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, https://doi.org/10.1007/978-3-030-77143-0_8 75
RkJQdWJsaXNoZXIy MTMzNzEzMQ==