An Assessment of Persistent Acoustic Monitoring of a Nuclear Reactor during Full Power Generation

Round 1

Reviewer 1 Report

This study uses smart phones recording low-frequency sound as a mean to monitor the HFIR reactor behavior at ORNL and demonstrates the record sound power (especially band-pass filtered RMS power) has a good correlation with reactor operation power. RedVox is easy to use on a smart phone but it seems very sensitive to ambient noise. Some suggestions to make this manuscript better: 

1) You used 10 smart phones but only show the data from 3 phones. How about the data on the other phones? Those phones give very noisy data like Phone 40, or show similar data as Phone 30 and Phone 35?

2) In additional to noise issue,  the Redvox shows some amount of latency between the acoustic data and reactor operation. If compared to other typical methods for low-frequency sound measurement, do you think the Redvox performance is acceptable?

3) Reactor radiation could cause ionization and further cause Low frequency EM waves. I suspect EM waves also after acoustic recording.  Did you do any control experiment on this?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Cardenas et al. demonstrates that stages of a nuclear power reactor operations can be monitoring with local distance (<1km from source) acoustic monitoring.  Instead of traditional infrasound sensors, the authors use low-cost novel sensors consisting of Samsung smartphones operating a RedVox app.  Sensors are deployed at increasing distances from the source.  Through examination of signal power and frequency characteristics, the authors demonstrate that signal power tracks reactor operations – by identifying a clear fundamental frequency that is consistent across all three sensors, the authors demonstrate that the primary acoustic source are fans in the facility.

Strengths:

The introduction and background sections are well written and introduce the problem nicely.  Prior relevant work is summarized in an accessible manner to the readers.  Reactor processes are explained well.  The methods to isolate signals using summed power spectral density are well explained and work well in this scenario. 

Weaknesses:

1.     I was expecting to see a map or illustration of the network design relative to the facility.  Reading the caption of Table 1 – how is phone 33 located 101 m from the towers but only 34 m from the facility itself.  The authors make the point to differentiate these distances but then do not discuss them in the text

2.     The RMS power figures and discussion of the three cycles are repetitive – I suggest include perhaps Figure 1 +3 and putting the rest into supplementary materials.  Similarly I suggest condensing explanations of each cycle to focus on results that were and were not similar across the cycles i.e. issues with fan noise during cycle 488

3.     The fundamental frequencies at each phone are not immediately evident looking at Figure 2 – this claim should be better substantiated with improvements to the figure.  Figure 11 illustrates these frequencies much better – perhaps moving this figure earlier in the paper?

4.     Phone 35 is only 40m further from the facility but exhibits much lower power, do you have an idea for why? Is there a way to extrapolate signal power as a function of distance?

5.     Results and conclusions indicate that the signals are sources from cooling tower fans, but are not substantiated. If you know the blade speeds and specifications, could you not calculate what these frequencies should be following examples from wind turbine papers similar to Li, J., Chen, J., & Chen, X. (2011). Dynamic characteristics analysis of the offshore wind turbine blades. Journal of Marine Science and Application, 10(1), 82-87.? This claim would be better supported by calculations rather that speculation. If the cooling tower fans are too different from wind turbines to make these calculations, disregard my question.

6.     Conclusions are missing statements regarding the use of smartphones over traditional infrasound sensors - given that Marcillo et al used more traditional sensors and your results are similar, this lays the groundwork to suggest that smartphones may be used in place of traditional sensors for a variety of reasons.  However, if you are going to make this claim I would like to see some further assessment of the noise characteristics - you mention noise as a source several times in the manuscript but do not go into details.  It would be interesting to compare noise from your sensors and how it obscures signals to noise/signals from the Marcillo study.

See attached annotated file for specific comments.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This was an interesting data correlation paper; the subject of monitoring nuclear power plants is timely and essential due to the Ukrainian crisis, and the standoff aspect of infrasound is appealing.  It is a validation and observation paper related to prior published work, which can be summarized as confirming previously observed signals from further distances. The work presented in this manuscript appears to be performed on data that was collected in an experiment that was not designed to test the hypothesis as written here, however. Robust infrasound experimentation has a generally accepted set of protocols, including meteorological measurements, estimation of predicted SPL relative to noise to select ideal sensor placement, notation of actual noise sources of note during experimentation, performance analysis on the microphones used in the study, and validation/verification of sources based on source models or direct collection of the source with on-source instrumentation. I have noted discrepancies in the experimental methodology in light of these accepted protocols, though it is not clear whether the lack in the manuscript was due to space constraints or whether they were not performed at all.

There are a few concerns related to the experimental design as described in this manuscript, but none of these are insurmountable provided the journal recognizes the limitations and desires publication. If the supporting observations or modeling have not been done, this reviewer acknowledges it would be too onerous to include in this manuscript, both in time and cost. If that is the case, making clear why they were not done, not included, or not relevant would suffice for edits. Most of these comments are tied to specific lines in the manuscript and denoted below.

One overall thing that is missing from the experimental design and protocol is the calculation of expected power degradation at that distance. This is a geometrical spherical spreading problem, at these distances and frequencies. There is no section that talks about modeling of expected power and signal amplitude loss at these distances, particularly related to the observed noise floor. The PSD measurements are interesting, but what would the expected values be? Given the observations at closer distances from prior reports, there could have been estimations of expected power at those ranges but none of this was included. That may be what drove the selection of the three microphones included, but without this information it’s difficult to justify some of the experimental/analytical choices. It’s likely that those choices to not include expected information were made deliberately and not considered to be within the scope or length requirements of the paper but are important for understanding the analysis and motivation of the study beyond ‘we saw what they saw’ related to the papers in the introduction section.  Tied into these estimates would be predicted attenuation curves, as well as understanding the component of turbulence to predicted signals, particularly at the 21.4 Hz expected frequency. This would add relevance to the claim that infrasound monitoring at standoff distances would be practical and beneficial for nuclear power plant performance.

Several specific comments follow, and should these recommendations be addressed, the manuscript would be acceptable for publication.

(Introduction, lines 62-73) There is repeated mention of ‘data fusion’. This generic term is not informative and should be explicitly stated what techniques, or at least what level fusion, were applied in these studies. Is this level 0? Level 1? Running or static? Was it a mix of contact and non-contact sensors? If this section is serving as a general lit review for the MINOS studies, it would be helpful to provide enough details to indicate whether this study could be included in future fusion efforts.

(intro, line 77) ‘Similar fashion’ is vague. Either the same techniques are employed, or they are not. What similarities? How would the variability from the initial study affect the results?

(Materials and Methods, line 113-114) Why these three phones? There is no indication for why all 10 are not considered, and without justification it leads this reviewer to ask if the data was included to support the hypothesis, not because it was the totality of the actual data (which may not have supported the hypothesis). Please state why this choice was made and what implications it has to the overall data analysis.

(M&M, table 1) Why was no map of the study area provided? If this is a sensitive area and that cannot be the case, why include the actual GPS lat/long of the reactor? The distance to the facility measurement does not take into account the directional propagation or other obstructions to recording equivalent data. While this map may very well exist in some of the cited work, it behooves the authors to provide a graph to properly orient the reader to the data. If a true map is not allowed, perhaps an illustrative layout would suffice. If building topography could be included, this would be beneficial, particularly related to the dimensions of the buildings. The terrain of built structures does have measurable impact on observable signatures in the frequency range analyzed here, so that is important data for the readers. However, this reviewer does understand location disclosure sensitivities, and would be satisfied with acknowledgement of such and an orientation along path. This would also help mitigate the potential that other industrial sources could produce either the signals or noise/clutter observed on the three stations analyzed in this paper. Related, what are the orientations of the sample microphones to any existing industrial or otherwise anthropomorphic clutter during the study and does it change between collects (ie a generator was operational near one microphone during only one collect, but not the others).

(M&M, lines 131-133) Why was this frequency band selected for analysis? What are the actual microphone specifications for the cell phones used? Were all phones in place for the totality of the experiment and were the phones moved at all during the duration of the experiment? Did any calibration occur? If yes, were they calibrated at the same time? If no, how does the performance drift affect the signals observed? Was this analysis done post-collection? What is the frequency degradation expected due to exposure to the elements? Were any physical filters employed? The manuscript does state that the phones are inside of structures that protect them from the elements, but not what the amplification/degradation of the signals relative to the frequencies of interest. What was the orientation of the microphone component of the cell phone relative to the enclosure mechanics? Is the input port for the microphone in-line with the shroud components (for lack of a better way to describe the housing), and is there an idea of how much that mechanical filtering would affect the observed signals here? If this is part of the existing MINOS studies, state and cite, please.

(Results and Discussion, general) There is no detailed discussion of meteorological data in this manuscript. Unfortunately, this is a confounding factor in the analysis of results. Was there a met station on site? What elevation was it in relation to the observation points? Were the wind speed/ direction and other turbulence effects taken into account in the analysis process? Were all collects under the same temperature, pressure, and wind conditions? How would this affect the observation? This reviewer doesn’t see that this analysis has been performed, and it would be well within the experimental protocols to record this information (or should have been). If the authors deem this necessary to exclude, please to include the justification.

(R&D line 193) What would that low frequency noise be? Without an area map, it is difficult for the reviewer to parse what that extra source would be.

(R&D, line 208-209) Explicit reference to thermal effects are made here, without data to back up the claim. Were meteorological conditions observed during this study?

(R&D, line 213) Mentions of wind reiterate the concerns of the comments above.

(R&D, Figure 2) The scale is distracting. Why not use the more commonly accepted spacing of 10, instead of 10.7 The scales on the vertical access are quite variable; this is acceptable. The breakdown of 10.7 is odd. The Figure 11 uses frequency band blocks that are acceptable, at 2 Hz, starting from 0. Something similar should be used in Figure 2.

(R&D, line 242-244) Without a discussion of potential other noise sources and their expected frequencies this claim is suspect. What other things that generate infrasound are in the area? If none, state so. This would be easily illustrated with a diagram or map overlay, see earlier comment.

(R&D, line 294-295) The line regarding the inability to verify the fan operation begs the question of modeling the fan performance in relation to expected acoustic emissions. This reviewer believes that this must have been beyond the scope of the work already performed or it would have been included. If the author team thinks this would be a good way to verify the work, state so. Future work is opportunity, but the authors should indicate a potential method to verify, even if it is in the future. COMSOL coupled structural and acoustical models would work well here.

(Conclusions, general) Correlating signal strength to distance would be more robust with meteorological measurements, as noted above. Reiterating why those are not included in this paper would be beneficial though, if possible, meteorological observations should be included.

(Concl., line 443-445) The conclusion that additional signal frequencies are correlated to fans, either in rotational speed and motor frequencies does not appear to be supported by either modeling efforts or direct recording of the potential sources of interest by a complementary microphone on that source (otherwise known as a traveler gauge). If this was performed, state so and provide validation information. If not, perhaps alluding to the fact that this would be the best to way to verify, but that it was unavailable for this study.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

This paper presents a method for an incremental improvement on the remote acoustic monitoring of the mechanical components of a nuclear plant. It is well structured and clearly outlines the method used to improve on the monitoring in the presence of ambient noise. In addition to the comments made on the attached annotated PDF file, I suggest the following:

 * "Data fusion" is used in the manuscript without and explanation of what it is. There are many ways to fuse together different data or data types, and it would be useful to briefly explain how data fusion is applied in the references cited, and whether the authors believe that the method presented also qualifies as a data fusion exemple. 

* The text mentions four cooling towers, supposedly each with their own fan. However, Table 1 lists a single distance to the colling towers. It would be useful to provide a sketch of the map of the layout, with the location of the fan or fans, the Smartphones, and the HFIR facility. 

* The discussion would benefit from an explanation of why the pumps inside the HFIR do not seem to provide a large enough acoustic signal to be monitored, contrary to the fan in the cooling towers. It would also be useful to mention any other technology which might help (or not) with their remote monitoring.  

* The discussion or conclusion would benefit from suggestions for the next steps to enhance the method or bring it closer to operational implementation. 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Some Figure numbers like Figure 5 are deleted as well. Need to check and restore the figure numbers. 

Reviewer 2 Report

My comments have been addressed; I recommend accepting for publication.