IoT-Based Intelligent Monitoring System Applying RNN
Round 1
Reviewer 1 Report
1. GRUand LSTM already apply in so many prediction areas. What is the novelty of this paper?
2. The structure and parameters of the GRU and LSTM should be described.
3. The information of the input data is not clear. what variables do you use? using moving windows or using single-time data as input? The information on the output is lacking. Do you predict the next 1 min of sensors?
4. the data structure and time interval are not clear.
5. if the model is only used to predict the next 1 min, why do we need this model because the time interval is too short? the variables change can be regarded as a linear change. Maybe just using a linear model has the same result.
Author Response
Response to Reviewer 1 Comments
To the Reviewers and Editor:
We are very pleased to have received your valuable comments regarding our manuscript entitled “An IoT-based intelligent monitoring system applying RNN”.
In the revision step, we have revised our manuscript according to the editor’s and reviewers’ comments, and our revisions are provided below.
Point 1: GRU and LSTM already apply in so many prediction areas. What is the novelty of this paper?
Response 1: The novelity of our paper is that we propose an ICT convergence framework to support the intelligent predictive maintenance of biobanking systems. Since bio resources and data used in the life sciences are basic infrastructure that supports research capabilities and research indicators, we designed an intelligent monitoring framework applying IoT and Recurrent Neural Network to improve the reliability of the biobanking system.
Point 2: The structure and parameters of the GRU and LSTM should be described.
Response 2: The structure and parameters of GRU and LSTM are described in detail.
Point 3: The information of the input data is not clear. what variables do you use? using moving windows or using single-time data as input? The information on the output is lacking. Do you predict the next 1 min of sensors?
Response 3: The description of the input data has been supplemented in section4.
Point 4: the data structure and time interval are not clear.
Response 4: The data structures and time intervals are described in detail in section4.
Point 5: if the model is only used to predict the next 1 min, why do we need this model because the time interval is too short? the variables change can be regarded as a linear change. Maybe just using a linear model has the same result.
Response 5: Predicting 1 minute is because sensor data is collected every minute. If the interval of sensor data is set to 5 minutes, 5 minutes later can be predicted. This is because, by comparing the predicted data with the actual sensed temperature data, if the actual sensed temperature data has a big difference from the predicted data, it can be treated as anomaly detection. The focus is on maintenance rather than forecasting. In this paper, RNN is applied for predictive maintenance of biobank and an IoT-based monitoring framework that supports automated intelligent monitoring is proposed.
We greatly appreciate your helpful comments, which we feel have improved our manuscript.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Review comments from the reviewer
Article title: An IoT-based intelligent monitoring system applying RNN Comments and Suggestions for Authors
This research presents a IoT and deep-learning-based approach for monitoring BioBank system using LSTM and GRU. The subject matter is interesting and valuable. The objective and quality of the work is appreciable and has merit to publish by the journal. I have a few comments below, that may help the authors to improve their work:
Review comments from the reviewer
- In objective it is mentioned that the proposed architecture includes hardware, predictive model, and Biobank Manager an UI module, but there is no description about that BioBank Manager module in this article. Please add clear explanation.
- Figure 2 is not clear. Kindly include clear image.
- In line number 143, it is mentioned “if the between the actual sensed temperature and the predicted temperature exceeds the threshold value, it is determined that the biobank is in an abnormal state”. Is the threshold is same across biobank or differs? How do you derive the threshold value?
- In context of line number 146, how to figure out the failure in each part of cryogenic freezer?
- In section 2 Background study, kindly include many recent papers on deep learning techniques that are used for monitoring. In literature study kindly avoid continuous citations. Eg [20-23].
- Explanation on the figure 3 is missing. Also detail the parameters w,h,x, U, V, and O used in figure 3.
- In Section 3, the details of terminologies like front and rear temperature, internal temperature are unclear. Kindly discuss the functionality of the equipment parts under consideration.
- The details (type, size, etc) of the sensors used in the experiment should be included.
- In figure 4, what is the role of main controller? Please explain because in later part of the article it is understood that the data is collected directly from each sensor.
- In Figure 5, the process flow is unclear. Connectivity between the various parts of figure is missing.
- In Figure 6, the equipment description is unclear. Kindly include clear image.
- Is there any difference between the configuration of the sensor used across various parts of the equipment’s? If so, kindly include the details.
- In Figure 7, include clear image. The sample data shown in the figure can be given as a separate table. (it’s a suggestion)
- Details about the dataset like source, total size, training, and testing data size is missing. The description about the input values and unit of temperature measure is missing. Kindly include.
- Explanation on proposed methodology in terms of LSTM and GRU, please add some more points.
Mathematical explanation of the model developed need to be included.
- In Section 4 – Experiments, the explanation on evaluation metrics to be included.
- In line number 290, it is stated “However, the GRU is known to be efficient in consuming computational resources”. Please justify.
- Separate section as Results and discussion can be included.
- In section 5 conclusion, in line number 302, it is stated “The experimental results showed that the accuracy of the two models was similar, and it was confirmed that overfitting did not occur”. Is that only reason to confirm non -occurrence of overfitting. Please justify.
- In general, all figures need clarity. Include citations for the images, if used from other external sources.
Author Response
Response to Reviewer 2 Comments
To the Reviewers and Editor:
We are very pleased to have received your valuable comments regarding our manuscript entitled “An IoT-based intelligent monitoring system applying RNN”.
In the revision step, we have revised our manuscript according to the editor’s and reviewers’ comments, and our revisions are provided below.
Point 1: In objective it is mentioned that the proposed architecture includes hardware, predictive model, and Biobank Manager an UI module, but there is no description about that BioBank Manager module in this article. Please add clear explanation.
Response 1: The part marked as Biobank Manager in figure 4 is the client part, and the manager can monitor the biobanking system in real time through the client UI/UX. BioBank Manager is not a module but a user who takes a role of managing to monitor biobank using user interface(UI).
Point 2: Figure 2 is not clear. Kindly include clear image.
Response 2: It has been replaced by a clear image. The positions of Fig. 2 and Fig. 3 have been swapped, and some contents have been modified accordingly.
Point 3: In line number 143, it is mentioned “if the between the actual sensed temperature and the predicted temperature exceeds the threshold value, it is determined that the biobank is in an abnormal state”. Is the threshold is same across biobank or differs? How do you derive the threshold value?
Response 3: Threshold can be set according to each biobank. In the experiment of this paper, learning was carried out by setting the threshold differently depending on the part to which the temperature sensor is attached. The threshold can be set by the Biobank administrator.
Point 4: In context of line number 146, how to figure out the failure in each part of cryogenic freezer?
Response 4: In the method of detecting the failure of each component, when the pattern of the temperature data over time has a large difference from the actual sensed data, it is possible to predict the occurrence of an abnormality of the component. If the difference between the data of the normal pattern predicted by GRU or LSTM and the value of the actual sensed data is greater than the threshold, it is determined that there is a possibility of failure and the safety of the Biobank can be checked.
Point 5: In section 2 Background study, kindly include many recent papers on deep learning techniques that are used for monitoring. In literature study kindly avoid continuous citations. Eg [20-23].
Response 5: Section 2 is modified to avoid continuous citations.
Point 6: Explanation on the figure 3 is missing. Also detail the parameters w,h,x, U, V, and O used in figure 3.
Response 6: Figure 3 was replaced by other image as figure 2 introduced in reference [21] and explanations were added.
Point 7: In Section 3, the details of terminologies like front and rear temperature, internal temperature are unclear. Kindly discuss the functionality of the equipment parts under consideration.
Response 7: Details of terms such as front-to-rear temperature, internal temperature, and descriptions of the functions of equipment parts have been supplemented.
Point 8: The details (type, size, etc) of the sensors used in the experiment should be included.
Response 8: The sensors are pt100 type temperature sensors and detailed specifications of sensors used are described.
Point 9: In figure 4, what is the role of main controller? Please explain because in later part of the article it is understood that the data is collected directly from each sensor.
Response 9: The role of the main controller is to collect data from each sensor and send it to the server. It is a component to carry out data transmission and reception with built-in IOT.
Point 10: In Figure 5, the process flow is unclear. Connectivity between the various parts of figure is missing.
Response 10: Figure 5 has been modified to show the flow of the process and the relations of the various parts.
Point 11: In Figure 6, the equipment description is unclear. Kindly include clear image.
Response 11: The details of the equipments in figure 6 have been described.
Point 12: Is there any difference between the configuration of the sensor used across various parts of the equipment’s? If so, kindly include the details.
Response 12: The sensors are all temperature sensors, so there is no difference in the configuration of the sensors. However, the attachment location of the sensor may be different for each freezer equipment, and it is best to install it on the side of multiple refrigerant pipes.Attached positions of temperature sensors were installed for each major part of the biobank.
Point 13: In Figure 7, include clear image. The sample data shown in the figure can be given as a separate table. (it’s a suggestion)
Response 13: Although the sample data was not organized in a separate table, the description of the data was supplemented.
Point 14: Details about the dataset like source, total size, training, and testing data size is missing.
The description about the input values and unit of temperature measure is missing. Kindly include.
Response 14: The description of the temperature measurement unit and details of the dataset used for training have been described.
Point 15: Explanation on proposed methodology in terms of LSTM and GRU, please add some more points. Mathematical explanation of the model developed need to be included.
Response 15: Detailed explanation of LSTM and GRU is described.
Point 16: In Section 4 – Experiments, the explanation on evaluation metrics to be included.
Response 16: In general, in machine learning, an evaluation matrix can be used. As an evaluation index in deep learning, prediction accuracy can be used as an evaluation index. In this paper, to evaluate the prediction accuracy, the accuracy was calculated by comparing the values of the predicted data and the actual sensed data and calculating the cases within the threshold range, and the figure 9 is a graph that summarizes this.
Point 17: In line number 290, it is stated “However, the GRU is known to be efficient in consuming computational resources”. Please justify.
Response 17: In references [20-24],have already been researched and citations have been added.
Point 18: Separate section as Results and discussion can be included.
Response 18: Results and discussion have been included in one section.
Point 19: In section 5 conclusion, in line number 302, it is stated “The experimental results showed that the accuracy of the two models was similar, and it was confirmed that overfitting did not occur”.
Is that only reason to confirm non -occurrence of overfitting. Please justify.
Response 19: That overfitting did not occur is a matter that can be confirmed in the loss function graph of each sensor, and it is described as such as can be confirmed in each loss function graph for 6 sensors.
Point 20:
In general, all figures need clarity. Include citations for the images, if used from other external sources.
Response 20: Figures have been modified clearly and replaced by clear one. It has been checked that figures from external sources are used with citations.
We greatly appreciate your helpful comments, which we feel have improved our manuscript.
Author Response File: Author Response.docx
Reviewer 3 Report
An IoT-based intelligent monitoring system applying RNN
Comments-
1) The experiment and discussion part is very weak. Kindly modify it.
2) Why specifically RNN is used? Justify.
3) Comparison with existing models required.
4) Conclusion is not correlating with abstract. Rewrite.
5) Information presented is a bit scattered. Refine
6) Remove all linguistic errors.
Author Response
To the Reviewers and Editor:
We are very pleased to have received your valuable comments regarding our manuscript entitled “An IoT-based intelligent monitoring system applying RNN”.
In the revision step, we have revised our manuscript according to the editor’s and reviewers’ comments, and our revisions are provided below.
Point 1: The experiment and discussion part is very weak. Kindly modify it.
Response 1: It has been modified by adding explanations for the experiment and discussion part. And detail description of the data used in experiments has been supplemented.
Point 2: Why specifically RNN is used? Justify.
Response 2: The reason for applying RNN is that the temperature data collected from each device in the Biobank is time series data. RNN is a deep learning technology optimized for time series data prediction.
Point 3: Comparison with existing models required.
Response 3: Deep learning models for time series data have already been confirmed that RNNs are optimized in previous studies. In this paper, an experiment was performed to compare and analyze the two models in order to select a model with good performance between the GRU model and LSTM model among RNNs.
Point 4: Conclusion is not correlating with abstract. Rewrite.
Response 4: The conclusion part has been revised.
Point 5: Information presented is a bit scattered. Refine
Response 5: Overall, the contents of the paper have been revised.
Point 6: Remove all linguistic errors.
Response 6: The linguistic errors have been revised.
We greatly appreciate your helpful comments, which we feel have improved our manuscript.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
1. Neural numbers of each model are still unclear.
2. According to the answer to Point 5, the authors think maintenance is what they focusing on, and explain if using 5 mins, the model is workable. However, the data information for 5 mins interval is different from the data information for 1 mins interval. they are not equivalent. The authors cannot just explain it without proving it. Furthermore, the maintenance should focus on Prognosis. The next 5 mins prediction is not enough. If the model predictions cannot be long enough to have enough reacting time for site operators, I really don't know what to do with this model.
Author Response
To the Reviewers and Editor:
We are very pleased to have received your valuable comments regarding our manuscript entitled “ An IoT-based intelligent monitoring system applying RNN”.
In the revision step, our title has been modified as follows: “IoT-based intelligent monitoring system applying RNN”, and we have revised our manuscript according to the editor’s and reviewers’ comments, and our revisions are provided below.
Our manuscript was revised by Editage Experts again to improve the language, and editing certificate is attached.
Point 1: Neural numbers of each model are still unclear.
Response 1: Our model is designed with two hidden layers in sequence 10.
Point 2: According to the answer to Point 5, the authors think maintenance is what they focusing on, and explain if using 5 mins, the model is workable. However, the data information for 5 mins interval is different from the data information for 1 mins interval. they are not equivalent. The authors cannot just explain it without proving it. Furthermore, the maintenance should focus on Prognosis. The next 5 mins prediction is not enough. If the model predictions cannot be long enough to have enough reacting time for site operators, I really don't know what to do with this model.
Response 2: What we mentioned in the response for point 5 is the description of the data used in the model built for the experiment. The 5 minute interval in previous response means it can be done if necessary. It is considered that the prediction interval may vary depending on the types of biobank or bio-resouces.
Since the temperature data is time series data, the temperature data pattern of the biobank can be predicted by the RNN training model. Although it is necessary to figure out the optimal interval according to the biobank types.
Our study suggests that it is possible to automate the detection of abnormal status of the biobank by attaching IoT devices to the biobank and applying RNN, and deep RNN can guarantee high accuracy. The proposed IoT-based intelligent monitoring framework applying RNN can be used to support the predictive maintenance of the biobank in practical use. As a future work, we will gather more data and perform additional experiments for practical use.
We greatly appreciate your valuable review comments, which we feel have improved our manuscript.
Author Response File: Author Response.docx
Reviewer 2 Report
Satisfied and can be considered for publication
Author Response
To the Reviewers and Editor:
We are very pleased to have received your valuable comments regarding our manuscript entitled “ An IoT-based intelligent monitoring system applying RNN”.
In the revision step, our title has been modified as follows: “IoT-based intelligent monitoring system applying RNN”, and we have revised our manuscript according to the editor’s and reviewers’ comments.
Our manuscript was revised by Editage Experts again to improve the language, and editing certificate is attached.
We greatly appreciate your valuable review comments, which we feel have improved our manuscript.
Author Response File: Author Response.docx
Reviewer 3 Report
Edits seem fine.
Manuscript can be considered
Author Response
To the Reviewers and Editor:
We are very pleased to have received your valuable comments regarding our manuscript entitled “ An IoT-based intelligent monitoring system applying RNN”.
In the revision step, our title has been modified as follows: “IoT-based intelligent monitoring system applying RNN”, and we have revised our manuscript according to the editor’s and reviewers’ comments.
Our manuscript was revised by Editage Experts again to improve the language, and editing certificate is attached.
We greatly appreciate your valuable review comments, which we feel have improved our manuscript.
Author Response File: Author Response.docx
