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

481

6

Fig. 4. Performance of the developed model: (a) full-time series, (b) selected time-window [6 - 7 sec].

3.2. Modelling and forecasting After the initial post-processing of the measured data a set of displacement data has been utilized to develop an autoregressive model i.e. AIR. The performance of the developed model has been cross-checked with the original data and it is clearly evident (see Fig. 4) that the developed model is capable of capturing the original behaviour quite accurately. In most cases, ones may argue that this is obvious as the model’s performance has been compared with the same data set that was used to develop the model. Hence, further, the aforementioned model’s performance has been validated with a different set of dis placement data. The validation results have been presented in Fig. 5. And interestingly, it is observed that the validation results also confirmed that the model is quite reliable in terms of capturing original dynamics of the signal. Also, it needs to be mentioned that the model was developed with 2-steps ahead prediction while the validation is done 5-steps ahead prediction. And still the accuracy of the validation is quite good e.g. 96.69%. In other words, if the validation was done for same as developed model (e.g. 2-steps), even with different set of data, the accuracy would be better ca. 99%. In the final stage, the model has been utilized to forecast the future unseen performance and depicted in Fig. 6. The results are quite good and such information might be very useful to have an idea what might comes next. However, this information doesn’t guarantee the outcomes are 100% accurate due to the underlying complexity of the model and data. 4. Conclusion This work has investigated the possibilities of developing a model via the use of measured time-series data based model that can be further implemented for forecasting unseen behaviour. The use and development of non-physics based model or data based models are on the rise due to the underlying benefits of such models. The aforementioned model can be useful not only for the modelling of the existing data dynamics but also such model can be used to predict future dynamics. Herein, an ARI type model has been utilized