Extruder Machine Gear Fault Detection Using Autoencoder LSTM via Sensor Fusion Approach
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe paper proposes an LSTM autoencoder-based mechanism for extruder machine gear fault detection. The paper is very interesting and generally well-written. However, some issues need to be solved:
1. The paper is full of typos (e.g. a) in the title of the paper is written “senor” instead of “sensor”; b) line 15 – there is an unwanted “)” there; c) many words are capitalized without a reason – see for example the lines 220-234; d) title of Table 3 contains “..”; etc.). Please carefully check the entire manuscript.
2. Please use either “auto-encoder” or “autoencoder” but not both!
3. Line 288: it is written: “Equation [].” Please correct.
4. Please present some reasons you selected an autoencoder LSTM and not another ML technique for fault detection.
5. Some figures have no labels on the x-axis. See for example Figs. 5a and 5b.
6. Many references (see the reference list) are not formatted as requested (e.g. references [40] or [45]) and references [45] and [46] are identical.
Comments on the Quality of English Language
The paper is full of typos that need to be corrected.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe paper reports an interesting study devoted to the defect detection in various gear mechanisms. Early defect detection is of crucial importance in heavy machinery lifecycle. The Authors propose a method of analyzing vibration and thermal data using LSTM model. The experimental results are convincing, and the study definitely deserves publication. However, I have several questions and concerns I want to address to the Authors. Please, find them below.
1. The Authors use a discrete wavelet decomposition for enhanced vibration signal analysis in their paper. Please, clarify the reasons of this choice. As far as this Reviewer knows, there are many efficient methods for solving vibration signal analysis task, e.g. enhanced discrete Fourier transform.
2. I believe the Authors overuse the capitalization of letters throughout the text.
3. The Authors use a separate sensor for vibration data analysis. However, for gearboxes used in electrical engines it is possible to detect the faults analyzing the rotor currents. Is your approach applicable to this case?
4. There are several recently developed approaches, that are applicable for machine fault detection, e.g. spiking neural networks and spectral markers. I recommend adding to the literature review\discussion the comparison between reservoir artificial and spiking neural networks in machine fault detection tasks and the suggested approach.
5. Tooth break is not the only possible fault in gearbox machines. The wearing of bearings and shaft spalling are very common issues. Can your approach detect these faults as well, or the system needs to be re-learned on new datasets? This matter is of great importance because competing techniques, e.g. spectral markers, are more generalized and less problem-specific detectors.
6. Fig. 5 (A) can be clearly distinguished from Fig.5 (B) without any wavelet decomposition. What about more tricky cases with slighter defects? If we are speaking about "early fault detection", it is well-known that defects are not very noticeable at the beginning.
6. I believe the plot shown in Fig. 9 is not very informative after Epoch 5.
Nevertheless, I like this comprehensive study and can recommend it for publication after proper revisions.
Comments on the Quality of English LanguageThe paper should be extensively proofreaded during the revision process. Please, pay special attention to punctuation, article usage and capitalization issues.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis paper studied the vibration and thermal datasets from two extruder machine gearboxes using an autoencoder Long Short-Term Memory (LSTM) model. It used a thorough global metrics evaluation methodology to further test the model’s dependability and efficacy.
Some of the equations were not properly referred to in the text, such as page 7, line 288, the wavelet transform is shown in Equation []?
The parameters in all formulas should be properly explained, including those in eqs.3-5.
The clarity in some figures is too low, especially the text within it. Increase the font size of x, y axes in figure 5, 7.
Fig. 10 is not clear. Increase the font size.
The introduction and literature review can be extended a bit to include studies on the extrusion technology, such as ‘Feasibility studies of a novel extrusion process for curved profiles: Experimentation and modelling’; ‘Analysis and modelling of a novel process for extruding curved metal alloy profiles’.
I suggest separating the discussion and conclusion sections, enrich the discussion part, such as including the strengths and limitations of current work, what is the future work, etc.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for Authors1) In the Abstract the Authors should place more exact results obtained in this research (Autoencoder LSTM accuracy rate only did not look fully convincing and sufficient).
2) Please avoid extensive usage of term “we” (for example at the end of Section 2). The paper is written and the research is performed by the Authors, so it is not necessary to highlight what “we” perform. The whole paper should be written in a neutral form, for example, instead of “we perform” should be used “it is performed”, etc. The corrections are required throughout the paper text.
3) Table 1 – selected Architecture Parameters should be better described. As the Authors have written before this Table, there are various Architecture Parameters which can be selected – so the combination of chosen ones should be properly described and explained.
4) Call on Table 2 is missing in the paper text.
5) Table 3 – obtained Accuracy and F1-Score are very high and it seems that this approach predictions are very satisfying. However, very high Accuracy and F1-Score are not sufficient by itself. The Authors should perform comparison with other similar models for the same or similar problems. Only such comparison can properly evaluate are the obtained Accuracy and F1-Score sufficiently high or not. I believe that this is very important and that should be included in the paper during the revision process – without mentioned comparison the numerical model Accuracy and F1-Score can be debatable (regardless of the fact that they are very high).
6) It will be helpful to any reader that inside the Nomenclature, along with the abbreviations, the Authors put and explain also all symbols and markings used in the paper, in the equations, etc. At the moment, paper reading can be problematic due to many symbols and markings explained in the paper text – to find a proper meaning the reader is required to turn back through the paper. If all the mentioned is placed and explained in one place (in the Nomenclature), the paper readability will be notably improved.
Final remarks: This is a very interesting paper and well performed experimental and numerical research. The obtained results are satisfactory in general, but they should be compared with other similar numerical model results (comment 5). However, all my above comments, except comment 5, are actually minor ones.
Comments on the Quality of English LanguageMinor editing of English language required.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors have successfully solved all my comments and concerns.
Reviewer 2 Report
Comments and Suggestions for AuthorsThank you for providing the revised version of your manuscript and point-by-point reply letter. Unfortunately, the Authors did not indicate changes in the reviewed manuscript, which complicates the comparison between versions. Nevertheless, I carefully checked the paper and believe it can be published with only minor fixes.
Comments on the Quality of English LanguageEnglish style and language are generally fine, just minor text polishing is needed.
