Description
Non-physics-based modeling is gaining attention in the area of structural health monitoring due to underlying advantages e.g. finite-element or physics-based model might not exist at all. Therefore, in case of aforementioned scenario, it may be useful to have a model that is developed via the use of time-series data instead of underlying physics. Such model can be developed using parametric, non-parametric, and system identification approach, where later state-space based model quantities can be derived. In this study, experimentally measured time-series data (e.g. displacement) have been utilized to develop a representative model. Initially, the data has been passed through a screening and selection process including data cleaning and filtering. Subsequently, the filtered data has been validated with the original data to make sure that overall dynamics remain same. Later, cleaned data have been employed to have an illustrative model via employing autoregressive type models. The influence of the model orders have been investigated and a comparison of different results have been presented. Further, the least-squares approach has been adopted to minimizes the prediction errors to optimize overall performance. However, it needs to be mentioned as herein out-only measured data has been used therefore no transfer function can be derived also stabilization plot is not possible without having a transfer function. Finally, the developed models output have been validated in both time and frequency domain. In short, it has been observed that the model output may vary significantly depending on the model orders and noise contamination of the original signal. The investigated approach is beneficial for monitoring while physics-based models either not available or not possible to derive.
