Text2shape Deep Retrieval Model: Generating Initial Cases for Mechanical Part Redesign under the Context of Case-Based Reasoning

Round 1

Reviewer 1 Report

Overall this paper is well written. However there are a few things I would like the authors to address before publish:

Would it be possible to use a more recent reference for Ref #1?

In the introduction the authors mentioned both methods have disadvantages yet only one was explained. In the next paragraph the authors again mention “for the problem above”. Please consider adding the justification for the simple reasoning algorithm.

Who provided the text description for either the unstructured or structured text description? It is also very interesting in figure 2 and line 227 the authors indicated design requirements expressed by customers…. Who is the customers? Are the end-user? From a design point of view, the customer (end user) will definitely NOT specify things like linking rod has two holes….they just want to have something fulfil their requirements. A more realistic expression would be “a component that connects to shafts whilst allowing free rotation” or even more generic like “enable free rotation”. If this is the case how will you implement the text2shape? Even taking a step back, in a product design specification the detail features would be specified. All you will find is some engineering requirements rather than detailed descriptions of geometric features. Therefore rationale behind this paper’s approach needs to be better justified.

Also, a more generic question, have you considered how your approach promote/hinder design innovation?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Overall the paper has been written very well and requires minor changes to improve.

Abstract

Consider revising the sentence or break into two sentences. A reader may find it challenging to understand what the author is trying to explain.

However, on 11 the one hand the manual retrieving method has the problem of low efficiency when the case base is 12 large, and on the other hand it is difficult for simple reasoning algorithms (e.g., rule-based reasoning, 13 decision tree) to cover all the features in complicated design solutions.

In the introduction part, the literature studies have been discussed and written well. A good number of references were included in the paper. Well done.

Section 3.2 mentioned there was a training set with 1000 samples established. Why was this 1000 chosen? Any specific reason?

The discussion and conclusion were clearly written. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

I have read the manuscript entitled “Text2shape deep retrieval model: Generating initial cases for mechanical part redesign under the context of case-based reasoning”. This study proposes a text2shape deep retrieval model to support text descriptions based on mechanical part shape retrieval, where the texts are for describing the structural features of the target mechanical parts. Although the structure of the paper is reasonable, the innovation of methods is insufficient and has not far reached the requirements of publication. There are still some questions as follows:

1. In Section 3.3.1, how to integrate CNN and RNN is briefly illustrated in Figure 6, but there is no detailed step description

2. In Section 4.2, how to preprocess the training set is not explained, and how the training set before and after the preprocessing was not reflected.

3. Each figure in the text contains too much information, and several sections of content are expressed in the same figure, so the figure is too miscellaneous and messy. For example, it is recommended to reduce the content of Figure 6 appropriately to avoid the repeated introduction of some basic concepts in RNN.

4. The English expressions in this paper need to pay attention to sentence breaks and reduce long expressions. And grammatical problems have affected the readers' normal reading and will cause misunderstandings.

5. CNN and RNN are very mature tools, the innovation of methods is insufficient and has not far reached the requirements of publication.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

The reviewer considered that the comments made were satisfactorily met and considered for publication.