An Improved Method Based on EEMD-LSTM to Predict Missing Measured Data of Structural Sensors

Round 1

Reviewer 1 Report

The authors present a study on predicting missing ‘dynamic response’ in structural analysis problems. The proposed method consists of two steps – i.e. EEMD and then a deep learning model.

 

Lines 64-68 are duplicated to 69-72.

It would be clearer if the cycle unit in figure 2 could be annotated in figure 3.

Line248 and figure 4, please annotate the pile. Similarly, in figure 5, please include the unit for the dimension and annotate a coordinate system.

In figure 5, are those triaxial accelerometers? There seem to be 11 sensors; why are only three measurements considered in the case study section? From which three sensors in figure 5?

Figure 11 (a) the legend blocks some of the BiGRU results.

Figure 12, please add an explanation of the five ‘shaded’ regions.

The results comparison in figure 11 shows an improvement of the proposed method to the ‘single method’. However, the difference between DNN BiGRU GRU and LSTM is barely distinguishable in figures 13 and 15. It is not quite apparent that LSTM is more accurate than the others, as mentioned in the conclusion.

 

The input of the feature layer of the deep learning models (DLM) can be application dependent. In this paper, IMFs are considered. The performance comparison is based on the available data only. There may be more comprehensive performance evaluations of the DLM in the literature. It would be better to include results from a physics-based model so that the performance of the DLM applied to structural dynamics can be rigorously evaluated. 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The article is interesting and well written. The shortcomings of the study were already highlighted in the conclusion. This article is a first of expected series. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

can the authors apply these techniques on a complete model like one high rise building and apply how can you find as example the missing data of the accelerations of some top points of the building.

Author Response

Please see the attachment.

Author Response File: Author Response.docx