PSI - Issue 64

Mohammad Shamim Miah et al. / Procedia Structural Integrity 64 (2024) 476–483 M.S. Miah and W. Lienhart / Structural Integrity Procedia 00 (2024) 000–000

477

2

1. Introduction The structural health monitoring is always linked with some sort of sensory information e.g. displacement, acceleration, strain, etc. The measured sensors information are either post-processed in real-time of offline to find out the useful feathers i.e. frequencies, mode shapes. It requires heavy computing power regardless the post-processing strategy online or offline depending on the desired situation. Further, most of the time it might not be so simple task to perform the measurement on-site due to operation disruption. However, to avoid the aforementioned problem a virtual model can be developed based on the measured data. And later that model can be validated and used to track any changes on the project. To do this, it is essential to develop a model from the tests data that may be capable of capturing true behaviour as accurate as possible. However this task may not be so straightforward due to the model’s parameters as they need to be optimized in order to realize a representative model. The interest of data-based modelling is growing quite fast due to many underlying advantages in moni toring. There are many research works can be found in the existing collection of literature. However, those research works are so diversified that doesn’t limit to any specific field and applications. More specifically, the data-based model and its applications are not only limited to civil engineering applications rather it can be found in various branches of science and engineering Swider et al. (2019); Cheng et al. (2020); Kumar et al. (2022). Among many existing data-based models and their applications, few can be mentioned such as; sparse approach Lederman et al. (2017), reduced kernel recursive least squares algorithm Zhou et al. (2017), autoregressive Miah and Lienhart (2021), multi-kernel regression model Wang et al. (2017), Fast unsupervised learning methods Entezami et al. (2020), hidden Markov models Wu et al. (2015), artificial neural network modeling Hossain et al. (2017), deep learning Kumar et al. (2024); Azimi et al. (2020). It is important to note that those models have been utilized for diverse purpose. For instance, Lederman et al. (2017) has adopted a sparse approach to monitor the tracks, Zhou et al. (2017) has utilized the reduced kernel recursive least squares algorithm to monitor and predict the behaviour of an aero-engine. The multi kernel regression model has been used by Wang et al. (2017) for forecasting wind power, Hossain et al. (2017) has employed artificial neural network to predict the roughness index of flexible pavements, Miah and Lienhart (2023) used acceleration data to develop a model via the use of sub-space method. Kumar et al. (2024) has used the deep learning to transform the simulation data into experimental one. While Azimi et al. (2020) has reviewed the data-driven monitoring and damage detection via the use of deep learning. The data-based model updating/modifying and system identification are also gaining attention and the utilizations are focused into various implementations. In this context, the subspace type system identification is used as a tool to identified features of a desired state space model Favoreel et al. (2000); Overschee and Moor (2012). In terms of existing model modification by using measured data, many works have been focused to develop and modify complex non-linear models alike e.g. Bouc-Wen model Kwok et al. (2017); Miah et al. (2015) for different applications. Khosravani et al. (2016) has investigated a comparison of energy prediction of a bioclimatic building by adopting neural network type model. Whereas, Svendsen et al. (2022) studied the experimental data-based monitoring and damage detection of a steel bridge. It can be summarized from the above discussion that the data-based modelling is gaining quite a lot attention in the area of structural health monitoring (SHM). However, it can’t be said that which model works better for which type of data such as displacement, acceleration. This has motivated to investigate the modelling and predicting behaviour by using measured displacement data. In summary, an autoregressive type model has been developed via the use of time-series data, later, the developed model’s outcome has been validated and used to predict an unseen scenario. The results show that the developed model is quite good that can render the true data and capable of forecasting the future with good accuracy. 2. Overview of the problem The investigation of this study is performed by using experimentally measured displacement data of a 2m long steel bridge in the laboratory. The sensor’s laser was focused around under the mid-span of the bridge. The displacement time-series data has been recorded with a sampling frequency of 312.5Hz. The used sensor