A Novel Algorithm to Detect Broken Bars in Induction Motors
and Douglas Bressan Riffel *Round 1
Reviewer 1 Report
Dear Authors,
I have read with interest your paper as it deals with an aspect that I have been working on and publishing for quite some time.
The most interesting aspect of it is that you analyze the failure of unloaded motors during start-up. Unfortunately, I have to say that the paper fails to demonstrate the full validity of the method, which I will try to explain below.
First, the introductory section needs to be significantly improved. In my opinion, and as a specialist in this field, there is no mention of the latest published work on the detection of this fault in transient regimes, which, in my opinion, already solves the problem posed by the authors. The use of time-frequency decomposition techniques, such as the Adaptive Slope Transform, already solves what the authors propose. Secondly, I don't quite understand the advantage of analyzing a motor at no load at low voltage when the fault can be detected when the motor is running in any type of regime, stationary or transient. In the introduction, some statements are made that, in my humble opinion, are not true. For example:
- “Wavelet resulted in the best technique for detecting broken bars”.
- “Analysing the literature, there is a lack in the statement of reliable severity index for broken bar detection and that …”.
Any expert in this area will tell you that these statements are false.
The introduction must be rewritten and updated with more recent publications. The introduction also fails in explaining the novelty of the author’s proposal.
The purposes of sections 2.1 and 2.2 are not clear.
The explanations of the experimental setup section are also poor. First, figure 4 does not permit to observe differences between the rotors. Important information is missing: sampling frequency, duration of the startups, etc. For me, the most interesting point is how the faults were simulated with the help of WEG manufacturer. But there is no information to appreciate this point. In this section, the meaning of some sentences is unclear. For example:
- “For this study, an isolated motor-generator was used …”
- “The tests were powered by 90V-….”
- “… a direct driven start-up current will starts between 7 and 10 ….”
Section 6 also needs important improvements and a better analysis. The meaning of lines from 193 to 196 is also confusing.
The authors cite the works by Riera-Guasp that describe the time evolution of the LSH energy in direct start-up. This explains why the filter range of 30 Hz provides the best results. This is obvious.
In page 8, it is said “digital Gaussian filter always has a spectral leakage”. Are you sure? Spectral leakage is usually related to the sampling frequency, spectral resolution, etc.
In the same page, there is another meaningless sentence: “The algorithm follows to find when and the amplitude of the each local maximum Teager energy”.
Figure 10 must be improved using the same range in the Y-axis in all figures to facilitate the comparison to the reader.
In this section, another sentence is confusing: “the fact that the time difference between the maximums increases with the increasing frequency during start-ups ….”. I really don’t understand this conclusion.
The last sentence of this section is also unintelligible.
The conclusions section also needs a major overhaul. You can’t tell that your algorithm is better than traditional methods if you have not given evidence of this. No results with other methods are presented in the paper to support this claim.
I also recommend a revision of the format, style, and English writing. For example, some figure are never referenced in the text.
Best regards.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors have done a very good analysis considering several broken rotor bars. It is a very interesting study. I have few questions.
1) Have you ever tried testing this at different slips? Does this algorithm apply for any induction machine including when it is acting as an induction generator? It is a known fact that construction wise, it is same for induction motor and induction generator. Just a curiosity to know how this algorithm is behaving when this induction machine is acting as a generator. The reason why I am asking is when this is fitted with wind energy system, I just wanted to know the compatibility.
2) Please include the circuit diagram of the induction motor with all the resistances and reactance parameters with the broken bars and without the broken bars. It will be sufficient if you could include the circuit diagram for one of the broken rotor bars so that it will be helpful for people to understand the difference in the variation of resistances and other parameters in the design pictorially. Please try to express them in pu.
Very good job authors. I am recommending your paper towards acceptance with major revisions.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Dear Authors,
First of all, I want to thank you for the effort to address all my questions and concerns in such a short time.
I really appreciate your work because now I understand your paper better and believe that has merits to be published.
Unfortunately, I will make more suggestions because I think you are not getting the most of your work.
I will try to summarize here my understanding of your work because you wrote in your answers relevant information that is not
on the manuscript.
Firstly, I say again that the detection and quantification of broken rotor bars in induction motors operating in an industrial facility at steady-state or transient regimes is already solved. I stand by my claim. But, your work is not about monitoring induction motors operating in an industrial facility. As you mentioned in your answers, your methodology is being used by WEG to monitor its products. I am sure it is used as a quality control of the rotor. In an assembly line, it will take a long time to test the motors loaded using any fault detection method already known and published. So, I guess that you designed a methodolgy to test those motors unloaded. This is the real and novel contribution of your work. It cannot be used in motors operating in an industrial facility, but it will be and it is useful in an assembly line or on a motor workshop where a brake or load is not available. I am afraid that this is not explained in the introduction and this is the main reason I rejected
your paper the first time. Therefore, my main suggestion is that you must explain this in the introduction.
I also have the following suggestions to improve your paper and make it more readable and understandable.
Section: Introduction
1. "The load must be sufficiently high in order to ...". This statement is not exactly true. Some recent papers deal with the problem of fault detection in motors operation at low-slip. Maybe, you should mention this in the introduction.
2. The newly added second half of the introduction is better than the previous one. I believe that you can improve it more with my above
mentioned ideas.
3. Emphasize the novelty of your work.
4. I think this is the first paper where the fault was not simulated by drilling holes in the cage. This is something relevant and interesting.
5. Finish the introduction explaining the structure of the rest of the paper.
Section: Experimental setup
Please, explain the meaing of condition states 4 and 5. The reader does not know what means that broken rotor bars are separated by 90º. What is this angle? Degrees are mechanical or geometrical or electrical?
What is the rotor bar number? Please, you should show a figure with one captured startup current.
Section: Algortihm development
I think the x-axis label in figures 5, 6 and 7 is wrong. I think it should be "sample number" or it is better if you use just "time" in seconds or miliseconds.
Section 5: Results
Figure 8 needs a significant improvement:
1. x-axis label is also incorrect. I would prefer to see time units as the time-difference between maximums is used as fault detection.
2. Number all six subfigures.
3. Remove labels as Rotor 1 to Rotor 6 and describe the fault condition at the bottom of each subfigure.
Table 1 can also be improved. In the first column I would replace "rotor" by "rotor fault state or condition".
I would also group the fault conditions and use a description with words, not numbers.
As the time difference between peaks is used as fault indicator, you should discuss how sampling frequency can affect this time-difference.
There is a typo or mistake in the formula of line 187 in page 9.
I am not convinced by the results showed in figure 8 for a partially broken rotor bar. The first peak is not present. Does this mean that there is not
energy related to the fault harmonic during the first half of the startup? This is very curious. It could be checked with the DWT, but the startup can be too short.
I recommend to update sections "benchmarking" and "conclusions" taking into account my other comments about the introduction and the results.
Best regards.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
The authors have done a good job and have satisfied all my queries. I am recommending the paper towards acceptance.
Author Response
We are very grateful to all of your time and efforts in reviewing our paper.
Best Regards,
Authors
