Nonlinear autoregressive with exogenous input neural network for structural damage detection under ambient vibration
Time-series method has become of interest in damage detection, particularly for automated and continuous structural health monitoring. In comparison to the commonly used method based on modal data, time-series method offers a straightforward application due to having no requirement for modal analysi...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | Thesis |
| 語言: | English |
| 出版: |
2020
|
| 主題: | |
| 在線閱讀: | http://eprints.utm.my/id/eprint/92426/1/SarehatiUmarPSKA2020.pdf.pdf |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 總結: | Time-series method has become of interest in damage detection, particularly for automated and continuous structural health monitoring. In comparison to the commonly used method based on modal data, time-series method offers a straightforward application due to having no requirement for modal analysis. Sensor clustering has been proven effective in improving the ability of time-series method to detect, locate and quantify damage. However, most of the applications rely on free vibration response that can be obtained directly by impact testing, which is difficult to practice for in-service structures, or indirectly by transforming the ambient vibration response. Therefore, a reliable method that allows direct utilisation of ambient vibration response for damage detection in structures without any data transformation is proposed in this study. The implementation of the proposed response-only method involves a three-stage procedure; (i) sensor clustering, (ii) time-series modelling and (iii) damage detection. Each sensor cluster is represented by a time-series model called nonlinear autoregressive with exogenous inputs (NARX) model, which is developed via artificial neural network (ANN) using undamaged acceleration data. The model is then utilised for predicting the damaged response and the difference between prediction errors is used to extract damage sensitive feature (DSF). The existence of uncertainties is addressed through setting up a damage threshold using several sets of undamaged data. The effectiveness of the method is demonstrated through a numerical slab model and experimental structures of reinforced concrete slabs and steel arches. It is found that the proposed structural damage detection approach based on NARX neural network is superior to linear ARX model as the approach is able to detect damage under ambient vibration. The results show that the highest predicted DSF corresponds to the location of damage and its value increases relatively with the severity of damage. Better damage detection is obtained when damage threshold is integrated into the proposed approach where the precision is increased by more than 24%. Overall, the proposed method is proven applicable to identify the existence, location and relative severity of structural damage under ambient vibration. |
|---|