Design of a Digital Twin for an Industrial Vacuum Process: A Predictive Maintenance Approach

Round 1

Reviewer 1 Report

This is an interesting paper that providing a summary of the condition monitoring of Induction Motor and Digital Twin. This paper proposes a predictive maintenance approach based on the establishment of digital twin model of Induction Motor.

There are several main contributions in this paper: First of all, this paper reviews current studies for condition monitoring of Induction Motor and Digital Twin. Then the digital twin model for Induction Motor is established with Simulink software. Last but not the least, the proposed diagnostic protocol is clarified and the effectiveness of the method is proven. In general, this paper should be revised carefully. I have the following suggestions that can be referred to highlight this paper.

1.         The first section only introduces the research on condition monitoring and Digital Twin, and does not directly introduce the research points and innovations of this article. It is recommended to summarize the Introduction of the article and point out the innovations of this paper.

2.         What’s the meaning of 400Δ in Table 2, please check the correctness of the symbols.

3.         The important mechanical part in the Figure 4. The physical system with label should be highlighted and explained.

4.         The data processing is an important part of Digital Twin, please add an introduction about data collection and transmission, and introduce several important parameters which is updated in digital twin model of the fan/motor system of Figure 8.

5.         Figures 14 and 15 suggest drawing with more contrasting colors and increasing tolerance zones to better display the simulation results.

6.         In the last part, adding multiple sets of experiments to calculate the accuracy of fault diagnosis will make the results more convincing

7.         In the conclusion part, please clarify the most important results of this paper, no need to repeat the establishment process of this paper, and please simplify the conclusion of this paper.

 Overall, the paper should be revised carefully. I believe that the paper can be accepted if the authors make a serious effort to account for all the comments provided. I deem this paper a Major Revision.

Author Response

Dear reviewers,

Thank you for your constructive comments, which enrich the scientific value of this manuscript. All these comments have been taken into consideration and have been appropriately added to the paper. Everything that is marked in red in the manuscript corresponds to the modifications that have been made, and the responses to each comment, individually and point by point, have been introduced in the following.

Note: All the texts in Italian style below have been added to the manuscript.

This article has been reviewed by three reviewers. So, we will start answering their comments one by one.

Starting with, Reviewer 1:

Comment (1): The first section only introduces the research on condition monitoring and Digital Twin, and does not directly introduce the research points and innovations of this article. It is recommended to summarize the Introduction of the article and point out the innovations of this paper.

Response (1): Based on this comment, the following points, which highlights the innovations of the paper, have been added to the article as appropriate:

• This paper proposes a condition monitoring of ventilation systems using the digital twin approach.
• The most common types of faults have been examined and considered in the proposed approach.
• We have created an alarm that can warn maintenance managers in the industry of the severity of each malfunction, based on a statistical study of a specific type of fault, and apply it to the rest.
• There is now a reference that allows for the state that the real system should be in, whether it is fault signatures or power consumption.
• This paper created a software version of the actual system that allows users to experience what they cannot do in reality.
• Experimental and simulations results prove the effectiveness of this developed technique.
• The use of this approach has been limited in this article to a specific type of industrial systems (ventilation systems), but with some modifications, it can be used with different framework.

Also, at the request of another reviewer, the literary review was expanded and some papers that are highly related to this topic were added to the introduction.

After having finished the clarification of these previous points, the introduction of the paper was modified and summarized to become the following:

1. Introduction

In industrial installations, ventilation systems are numerous and increasingly require continuous monitoring. Any sudden failure of these components can result in significant damage. Condition Monitoring (CM) is a growing technology that enables the identification of incipient defects. It can prevent unexpected failure of critical elements, increasing the life of components while decreasing maintenance downtime and costs. These systems mainly consist of an electric motor, a fan, ducts, bearings, transmission shafts, etc. \cite{j1}. In most cases, the fan is not connected directly to the motor, mainly to achieve two objectives. The first one is to protect the motor since the fan is exposed to environmental disturbances and withstands different operating conditions. While the second goal is to control the rotation speed and torque of the fan by adjusting the diameter of gears or pulleys \cite{j2}. In summary, it can be said that this type of system can be divided into two main parts. The first is the electric motor, which we will consider, in this article, as an Induction Motor (IM) since it is the most commonly used in industrial sectors, including ventilation and suction systems \cite{j3}. The second part is the purely mechanical system that includes the rest of the components. So, in this article, we will deal with these systems as fan/motor systems.

The IM is subjected to several defects, the main of them can be classified as shown in Table \ref{table1}, where each malfunction and its type are listed, along with the work environment of each motor. As for the fan part, it can be affected by various faults, like any rotating system, but the most significant are: fan imbalance \cite{j4}, shafts misalignment \cite{j3}, belt defects \cite{j6}, and last but not least, bearing faults \cite{j7}.

Table 1 (Not modified)

These faults need to be predicted at an early stage, and appropriate actions are taken to ensure the smooth running of the system. There are several methods to achieve this goal, but vibration analysis is the most popular method for condition monitoring. Acoustic monitoring is a viable option because vibration creates acoustic noise. However, as these methods are expensive due to the additional transducers required, they are only appropriate for large machinery or highly critical applications such as wind turbines \cite{b31}, milling and grinding processes \cite{b170}. All system malfunctions affect the IM, either directly or indirectly, so Motor Current Signature Analysis (MCSA) can be employed for CM. This method is part of the general topic called Electrical Analysis, which has proven to be a very efficient and practical approach since the current and voltage signals are easy to monitor. This type of signal reduces the number of transducers installed since current and voltage transducers are already installed for control, safety, energy, and other reasons. Electrical signals have been used to detect and locate not only electrical faults but also mechanical faults in the system components \cite{b106}. MCSA has undergone significant development in the last 30 years \cite{b55}. Numerous studies and works have been reported, such as the book by Saad et al. \cite{b34} and the one by Thomson \cite{b172} that discuss CM using electrical signature analysis. All these advantages and the effectiveness of the MCSA prompted to adopt this approach in this research. Based on the above, and the fact that this article takes into account the evolution of industrial sectors, with today's fourth industrial revolution (industry 4.0), which does not require human beings, production can be done with cyber-physical links with much less error and high reliability. With this revolution, products are becoming capable of sensing their own condition and that of their environment, coupled with the ability to process and communicate this data, enables the development of digital twins that can simulate physical system operation, track energy consumption (monitoring electrical signatures), detect system faults (electrical, mechanical, vibratory), and allow experimentation that is hard done with actual installations \cite{j30}.

Since the concept was first coined by John Vickers and Michael Grieves \cite{j23}, many authors have attempted to define the term Digital Twin, beginning with the aerospace industry \cite{j24}, focusing on structural mechanics, materials science, and long-term performance prediction of air and spacecraft \cite{j25}. With the growth of Industry 4.0, the focus has shifted to manufacturing and smart products \cite{j24}. In this context, the digital twin can contribute to information continuity throughout the product life cycle \cite{j26,j27}, virtual commissioning of (manufacturing) systems \cite{j28}, and decision support and predictions of system behavior in the product development phase and in all subsequent phases of the life cycle on the basis of computer-aided implementations \cite{j29}. The digital twin as a concept contains three major parts, which are the physical objects in the real space, the virtual objects in the virtual space, and the data and information connections that linking the virtual and real products \cite{j32}.

The concept of the numerical model has been adopted in many previous kinds of research, and it has given satisfactory results in achieving its objectives. For example, but not limited, to study the dynamics of a belt-pulley-shaft system, its numerical model was developed by \cite{j8}. This concept was used also by \cite{j9} to monitor an electrical submersible pump toward mechanical seal failure. The investigation of the accelerating transient vibrations of a rotor system was done by \cite{j10}. The numerical model was utilized by \cite{j11} to monitor an aero-engine dual-rotor system with a fan blade out. The digital twin has many uses, even simulating the operation of cars on the road \cite{j12} or CNC machine tool \cite{j13} etc.

After reviewing some of applications of the digital twin, we move on to the content of this paper, where we use this concept for the condition monitoring of a ventilation system. The most common types of faults have been examined and considered in the proposed approach, in order to create an alarm that can warn maintenance managers in the industry of the severity of each malfunction, based on a statistical study of a specific type of fault, and apply it to the other defects. So, there is now a reference that allows for the state that the physical system should be in, whether the fault signatures or power consumption. In general, we created a software version of the actual system that allows users to experience what they cannot do in reality. Experimental and simulations results prove the effectiveness of this developed technique. The use of this approach has been limited in this article to a specific type of industrial systems (ventilation systems), but with some modifications, it can be used with different framework.

The uses of the digital twin have been mentioned generally in the above. Now we will switch to what is related to the topic of this research and discuss some of the research that has used this approach in the area of fault diagnosis. The concept was used by \cite{j9} to monitor an electric submersible pump towards a mechanical seal failure, and get good results. The paper published by \cite{ert1} proposes an intelligent fault diagnosis method for a triplex pump based on digital twin and deep transfer learning. The proposed method has been experimentally validated and gives satisfactory results. Another study, \cite{ert2}, was inspired by the fact that transfer diagnosis scenarios are limited to the experimental domain, the inter-domain marginal distribution and conditional distribution are difficult to simultaneously align, and each source-domain sample is assigned equal importance during the domain adaptation process, so it proposed a new joint transfer network for unsupervised bearing fault diagnosis from the simulation domain to the operational field. The digital twin approach was used by \cite{ert3} in the intelligent manufacturing area to monitor the health of a rotating machine. The main challenge of this study was the nonlinear dynamics and uncertainty involved in the machine degradation process. It successfully overcame this challenge and achieved the desired goal, as evidenced by the results.

This article is organized into four main sections. Section 2 describes the physical system. Then deals with the development of the digital twin. The conditional maintenance technique used in this research is outlined in it. After that, the re-sampled signal is exploited by a statistical diagnostic procedure. The obtained experimental and simulation results are presented in section 3 and discussed. The last section is the conclusion of this paper with a portal for future work.

Comment (2): What’s the meaning of 400Δ in Table 2, please check the correctness of the symbols.

Response (2): In the three-phase machine (Like the Induction Motor used in this paper) there are two types of connection, of these power supply phases:

• Delta (Δ) connection, as our case.
• Y connection.

So, the “400Δ” used in Table 2 means that the power supply utilized is three-phase voltage source (AC) delta connected.

To avoid confusion, the symbol (Δ) in the paper has been changed and replaced with the word “Delta”, and a column was added to the table that shows the type of connection of the supply voltage source is three-phase.

Comment (3): The important mechanical part in the Figure 4. The physical system with label should be highlighted and explained.

Response (3): Based on this comment, Section 2. 1. was modified to provide more explanation of the physical system.

So, this section became as follows:

2.1. Description of the physical system.

We had previously indicated that the system being dealt with is divided into two parts, the electric motor (IM) and the mechanical part (the fan and the transmission system). This system mainly consists of the Pulley 1 (marked “a”) connected directly to the motor. The power is transmitted from the motor to the Pulley 2 (marked “b”) by three belts (marked “c”). where each belt has a length (L) equal to 2300 mm type SPB Vbelt. The fan (marked “f”) is connected to the Pulley 2 by a transmission shaft (marked “e”), has a diameter of 45 mm and 80 mm in length. The diameter of pulleys 1 ($R_{p1}$) and 2 ($R_{p2}$) is 160 and 225 mm respectively. The two bearings (marked “d”) that support this shaft are type 22210 EK. The fan has eight blades that are made of steel 60A, like the rest components. Other parts (labeled from “1” to “12”) have little effect on the basic function of the system, so we ignored their influence during the modeling process. The label of each part is given in the appendix A at the end of the paper.

Comment (4): The data processing is an important part of Digital Twin, please add an introduction about data collection and transmission, and introduce several important parameters which is updated in digital twin model of the fan/motor system of Figure 8.

Response (4): In the first version of this paper, Figure 8 (Figure 7 in the revised version) was included without a full explanation, but your comment alerted us to this matter and the need to include an explanation.

Therefore, the following has been added to the manuscript as appropriate, as an explanation of the mentioned Figure.

The digital twin provides condition monitoring administrations in light of Simulink models created by real systems and information acquired in the real world. To develop a computerized model to accurately and continuously monitor the state of the real framework, it is important to examine a methodology for constant updating between the computerized model and the real system. The current system collects information in a bottom-up manner. It first acquires different information about the subsystem boundaries and then collects information about the perception at the frame level. Finally, it combines this information from the subsystems to the framework as a mark of well-being of the entire framework. This granular perspective works with the resulting foundation of the Digital Twin model in light of the actual components. In addition, the physical world needs to transfer many kinds of information to the digital environment, such as operating conditions, working situations, sensor records, and more. The physical system needs standard data communicational schemes to accomplish a uniform conversion of the various communication protocols or interfaces and a standard packaging of the information. Through these data communication features, the multi-type and multi-scale information is standardized, cleaned and packaged by the physical system, and then loaded to the Digital Twin model in the virtual world. This significantly improves the performance of the twin.

In our case, we only used current and voltage sensors, since we use MCSA technology. These sensors provide us with the real voltage and current signals. By analyzing any of these signals using the methods that will be explained later, we notice that they contain certain frequencies, and each one belongs to a specific part of the system. These frequencies are our target, we monitor them and update the signal received from the digital twin to contain the same frequencies and values, so as to obtain an iterative convergence between the measured and the calculated response, as illustrated in Figure 7, and as shown in the results achieved in this study.

Comment (5): Figures 14 and 15 suggest drawing with more contrasting colors and increasing tolerance zones to better display the simulation results.

Response (5): The two figures referenced in this commentary have been modified, and drawn with more contrasting colors, likewise the tolerance areas have been increased in Figure 15 (Figure 14 in the revised version) to better display the simulation results.

We also made an adjustment to Figures 10-17 (Figures 9-16 in the revised version) at the request of another reviewer, so that the font size and style remain consistent across all figures.

Comment (6): In the last part, adding multiple sets of experiments to calculate the accuracy of fault diagnosis will make the results more convincing.

Response (6): In the last section of the manuscript, statistical studies were used to determine the severity of the defect. The reviewer pointed out in his commentary an important point that helps to enrich the results of the manuscript, namely the addition of several sets of experiments to calculate the accuracy of fault diagnosis. In this study, we have dealt with one type of malfunction and generalized what applies to it to the others, since there is no specific approach that can be relied upon to deal with each malfunction, except for the broken rotor bar fault, the severity of which is determined based on a reference mentioned in the manuscript. And since the system studied in this research is a system that exists in an industry and performs its function, not an experimental device in a laboratory, we were limited in the ability to conduct experiments on it, so we were content with what was included in the manuscript. But this commentary can be seen as a starting point for future research and work.

Comment (7): In the conclusion part, please clarify the most important results of this paper, no need to repeat the establishment process of this paper, and please simplify the conclusion of this paper.

Response (7): To achieve what is requested in this comment, the conclusion has been reworded, simplified and avoiding repetition of the steps in the production of this paper, with clarification of the most important results. The presentation of future work was also expanded at the request of another reviewer, so the conclusion became the following:

4. Conclusion

This article discusses the topic of condition monitoring in ventilation systems based on the digital twin approach. The most common defects in this type of systems have been reviewed. The method adopted in the diagnostic process is the MCSA, where the Concordia method and the FFT analysis are used to process the current signal. An alarm has been created to determine the severity of the defects. We have now a reference that allows for the state that the real system should be in. Experimental and simulations results prove the effectiveness of this developed technique. The use of this approach has been limited to ventilation systems, but with some modifications, it can be used with different framework.

Finally, as a proposal of future works, the developed approach can be used with other types of industrial systems and different kinds of defects. Other techniques can also be adopted in processing the current signals. As well another topic of research, is the implementation of the artificial intelligence, like Artificial Neural Networks, Fuzzy and Neuro-Fuzzy logic, to create the alarm that determine the severity of the defects instead of the statistical study. This study deals with the system that operates at fixed speeds, in order to improve the efficiency of the system to handle different operating environments, it is possible to implement signal processing algorithms (such as adaptive notch filters and adaptive observer approach) that can track and estimate transient frequencies to detect faults in variable speed conditions.

Reviewer 2 Report

The authors address an interesting hot topic nowadays, which is the digital twin concept. They make use of this concept to monitor and detect faults in an industrial application. The most common types of defects were reviewed and considered in the proposed approach, i.e. monitoring a ventilation system. At the end, the authors create a sort of alarm threshold based on statistical method.

This important automatic alarm does not convince me yet as well as the approach to simulate a fault on the system.

Major comments must be addressed correctly to accept the paper:

1) The fault is a random phenomenon in nature. The authors agree that they consider ‘defect signature’ as a random variable. However, they create the fault on the transmission belts by a white noise, which may not represent the reality. Is the fault a stationary random process? Is it a random variable that follows a Gaussian distribution?

2) The authors show that the fault is not Gaussian distributed, see the histogram in Figure 16. However, the statistical method in section 2.4. using this assumption.

Based on above points, section 2.2.2. and 2.4. need to be improved accordingly.

Despite the major comments mentioned above, it is worth to highlight that digital twin concept for industrial applications is an important topic for industry today. This work is therefore of immediate interest to the community.

Author Response

Dear reviewers,

Thank you for your constructive comments, which enrich the scientific value of this manuscript. All these comments have been taken into consideration and have been appropriately added to the paper. Everything that is marked in red in the manuscript corresponds to the modifications that have been made, and the responses to each comment, individually and point by point, have been introduced in the following.

Note: All the texts in Italian style below have been added to the manuscript.

This article has been reviewed by three reviewers. So, we will start answering their comments one by one.

Reviewer 2:

Comment (1): The fault is a random phenomenon in nature. The authors agree that they consider ‘defect signature’ as a random variable. However, they create the fault on the transmission belts by a white noise, which may not represent the reality. Is the fault a stationary random process? Is it a random variable that follows a Gaussian distribution?

Response (1): In our study, we relied on an operating system that performs its function in the factory, not on a test bench in a laboratory, so we were very limited when dealing with it, which would mean that we could not actually create faults, therefore we resorted to simulation in order to achieve this purpose.

It should be noted that the signature of a fault is a frequency that appears at a certain value when processing the current signal, so after choosing a specific fault to deal with (belt defect), we started to use the Simulink program to create this kind of fault.

In reality, in most cases, the frequency of the fault is always present, but with a relatively low value, as has been indicated in the manuscript, then when a malfunction occurs, the value of this frequency begins to increase gradually, with the increase of the fault severity, until reaching a threshold in which it can be considered a serious malfunction and the necessary measures must be taken to avoid it.

The fault has been created using a random variable and its mean value and the degree of variation around this value have been set, so that it follows the Gaussian distribution. If we want to compare this case with reality, it does not exist, but it has fulfilled the desired function, because the objective was to create frequencies with a value higher than the value at which the malfunction alarm is triggered. Thus, when any malfunction occurs, the value of the frequency will be calculated and compared to the threshold, whether the malfunction is variable and follows a Gaussian distribution or not. Therefore, based on the above, the approach taken in this manuscript can be applied in real situations.

Comment (2): The authors show that the fault is not Gaussian distributed, see the histogram in Figure 16. However, the statistical method in section 2.4. using this assumption. Based on above points, section 2.2.2. and 2.4. need to be improved accordingly.

Response (2): The two sections mentioned have been modified based on this comment. So, the following has been added to the section 2.2.2.:

Since we are dealing with a system functioning in the industry, the possibility of creating defects is actually not possible, so we will use simulation to achieve this goal. Equation (\ref{td}) indicates that the frequency of the defect changes according to its type as shown in Table \ref{table4}. As for the severity of this defect, it is related to the variable A. Therefore, in order to change the severity of the defects we will create, we will use a random variable that will represent the value of A, where we will set its mean value and the degree of change about it.

As for section 2.4. it only discusses the approach that was taken, independently of the results. It shows the equations that were used to determine the threshold based only on the measured data and not on the simulation. In other words, what is mentioned in this section is only related to the creation of the alarm. Therefore, after the previous clarifications, there is no issue with keeping this part as is.  

Reviewer 3 Report

The author proposed condition monitoring of ventilation systems through the digital twin approach, where the Concordia method and the Fast Fourier Transform analysis are used to process the current signal, and physical and numerical system current measurements are obtained and compared. Generally speaking, this manuscript is well written and organized. The topic is interesting and has practical contribution, and the experimental validation is also detailed and convinced, however, this paper has several issues which should be properly addressed. Detailed comments can be found as follows:

(1) Is the proposed method sensitive to noise and load changes?

(2) It would be better to give an illustration the overall framework for the proposed model in Figure 8.

(3) Will Concordia method achieve better results when combined with other signal preprocessing methods, such as wavelet?

(4) It would be better to remain the font size and style consistent on all figures. For example, Figure 10-Figure 17.

(5) Consider expanding conclusions from this work and especially, description of future work.

(6) The literature review is not enough, please add some papers that are highly related to this topic, such as "Intelligent Fault Diagnosis of Machinery Using Digital Twin-assisted Deep Transfer Learning[J]. Reliability Engineering & System Safety, 2021", and “Novel joint transfer network for unsupervised bearing fault diagnosis from simulation domain to experimental domain[J]. IEEE-ASME Transactions on Mechatronics, 2022”

Author Response

Dear reviewers,

Thank you for your constructive comments, which enrich the scientific value of this manuscript. All these comments have been taken into consideration and have been appropriately added to the paper. Everything that is marked in red in the manuscript corresponds to the modifications that have been made, and the responses to each comment, individually and point by point, have been introduced in the following.

Note: All the texts in Italian style below have been added to the manuscript.

This article has been reviewed by three reviewers. So, we will start answering their comments one by one.

Reviewer 3:

Comment (1): Is the proposed method sensitive to noise and load changes?

Response (1): Regarding the noise, the proposed method relies on the observation of the amplitude of certain frequencies belonging to the components of the system. And through a statistical study, a certain limit has been created from which the malfunction can be considered serious and the necessary measures must be taken, as these frequencies are always present when processing the current signal, and their existence does not necessarily mean that there is a malfunction in the system, as stated in the paper. As long as we know the frequency of each fault based on the references mentioned, we can differentiate between the noise and the defect frequency, because the amplitude of the noise is known according to each signal, the operating environment and the sensor used, any increase in the amplitude of the frequency related to the defect, can be detected, which makes this method effective even in the presence of noise.

As for the change in load, the techniques adopted to process the current signal (Fast Fourier Transform) can only work at constant speed, independent of the load. But if the load causes a continuous change in speed, then this method loses its effectiveness, so we have put this point on the schedule for future work. Therefore, we will use other signal processing algorithms (such as adaptive notch filters and the adaptive observer approach) that can track and estimate transient frequencies to detect faults under variable speed conditions.

Comment (2): It would be better to give an illustration the overall framework for the proposed model in Figure 8.

Response (2): Based on this comment, section 2.2.3. of the document, which includes Figure 8 (Figure 7 in the revised version), has been modified accordingly. In addition, as indicated by another reviewer, an introduction on data collection and transmission has been added.  The various important parameters that are updated in the numerical twin model of the fan/motor system have also been presented.

Figure 7 shows, under the title Digital Twin in the blue part, the stages of building the digital twin, which begins with understanding and studying the physical system, its components and geometries, and then moving to the modeling stage by creating the free body diagram, then the use of the Newton's second law of motions to get the system equations. As, the digital twin provides condition monitoring administrations in light of Simulink models created by real systems and information acquired in the real world. To develop a computerized model to accurately and continuously monitor the state of the real framework, it is important to examine a methodology for constant updating between the computerized model and the real system. The current system collects information in a bottom-up manner. It first acquires different information about the subsystem boundaries and then collects information about the perception at the frame level. Finally, it combines this information from the subsystems to the framework as a mark of well-being of the entire framework. This granular perspective works with the resulting foundation of the Digital Twin model in light of the actual components. In addition, the physical world needs to transfer many kinds of information to the digital environment, such as operating conditions, working situations, sensor records, and more. The physical system needs standard data communicational schemes to accomplish a uniform conversion of the various communication protocols or interfaces and a standard packaging of the information. Through these data communication features, the multi-type and multi-scale information is standardized, cleaned and packaged by the physical system, and then loaded to the Digital Twin model in the virtual world. This significantly improves the performance of the twin.

In our case, we only used current and voltage sensors, since we use MCSA technology. These sensors provide us with the real voltage and current signals. By analyzing any of these signals using the methods that will be explained later, we notice that they contain certain frequencies, and each one belongs to a specific part of the system. These frequencies are our target, we monitor them and update the signal received from the digital twin to contain the same frequencies and values, so as to obtain an iterative convergence between the measured and the calculated response, as illustrated in Figure 7, and demonstrated in the results achieved in this study.

Comment (3): Will Concordia method achieve better results when combined with other signal preprocessing methods, such as wavelet?

Response (3): In this paper, the fast Fourier transform was adopted to process the modulated current signal obtained after applying the Concordia transform.

In this Comment, the reviewer has raised an important point, namely the use of other signal processing methods, such as wavelets.

The comparison between the FFT and the wavelet shows the superiority of the latter, in terms of preserving signal information in the time domain as well.

The classical Fourier transform, used in this research, of a function, indicates accurately the magnitude of the frequency component. But in doing so, the spatial duration is lost: there is no way of determining at what moment in time the signal is emitted. On the contrary, the wavelet technique provides information on the temporal range of the signal.

Therefore, the adopted method loses its effectiveness when dealing with time-varying signals, but since the processed system operates at constant speeds, the FFT shows its effectiveness in achieving the desired goal. In addition to the above and the simplicity of this method compared to other signal processing techniques, we decided to rely on it in this manuscript.

This comment has been taken into account, as a point to work on in future studies. As stated in the conclusion of this paper, in terms of adopting other signal processing techniques that can work with time varying signals.

Comment (4): It would be better to remain the font size and style consistent on all figures. For example, Figure 10-Figure 17.

Response (4): Based on this comment, some adjustments to Figures 10-17 (Figures 9-16 in the revised version) have been made, so that the font size and style remain consistent across all figures.

At the request of another reviewer, Figure 14 and 15 (Figure 13 and 14 in the revised version) have been modified, and drawn with more contrasting colors, likewise the tolerance areas have been increased in Figure 15 (Figure 14 in the revised version) to better display the simulation results.

Comment (5): Consider expanding conclusions from this work and especially, description of future work.

Response (5): The conclusion has been reformulated, so the future works and their description have been expanded, as requested in this comment.

Also, based on another reviewer comment, the conclusion has been simplified and avoiding repetition of the steps in the production of this paper, with clarification of the most important results, so the conclusion became the following:

4. Conclusion:

This article discusses the topic of condition monitoring in ventilation systems based on the digital twin approach. The most common defects in this type of systems have been reviewed. The method adopted in the diagnostic process is the MCSA, where the Concordia method and the FFT analysis are used to process the current signal. An alarm has been created to determine the severity of the defects. We have now a reference that allows for the state that the real system should be in. Experimental and simulations results prove the effectiveness of this developed technique. The use of this approach has been limited to ventilation systems, but with some modifications, it can be used with different framework.

Finally, as a proposal of future works, the developed approach can be used with other types of industrial systems and different kinds of defects. Other techniques can also be adopted in processing the current signals. As well another topic of research, is the implementation of the artificial intelligence, like Artificial Neural Networks, Fuzzy and Neuro-Fuzzy logic, to create the alarm that determine the severity of the defects instead of the statistical study. This study deals with the system that operates at fixed speeds, in order to improve the efficiency of the system to handle different operating environments, it is possible to implement signal processing algorithms (such as adaptive notch filters and adaptive observer approach) that can track and estimate transient frequencies to detect faults in variable speed conditions.

Comment (6): The literature review is not enough, please add some papers that are highly related to this topic, such as "Intelligent Fault Diagnosis of Machinery Using Digital Twin-assisted Deep Transfer Learning[J]. Reliability Engineering & System Safety, 2021", and “Novel joint transfer network for unsupervised bearing fault diagnosis from simulation domain to experimental domain[J]. IEEEASME Transactions on Mechatronics, 2022”

Response (6): The reviewer indicated in this comment that the literature review was not sufficient and suggested adding a few articles that address this topic, so the following text was added to the introduction of the paper appropriately:

The uses of the digital twin have been mentioned generally in the above. Now we will switch to what is related to the topic of this research and discuss some of the research that has used this approach in the area of fault diagnosis. The concept was used by \cite{j9} to monitor an electric submersible pump towards a mechanical seal failure, and get good results. The paper published by \cite{ert1} proposes an intelligent fault diagnosis method for a triplex pump based on digital twin and deep transfer learning. The proposed method has been experimentally validated and gives satisfactory results. Another study, \cite{ert2}, was inspired by the fact that transfer diagnosis scenarios are limited to the experimental domain, the inter-domain marginal distribution and conditional distribution are difficult to simultaneously align, and each source-domain sample is assigned equal importance during the domain adaptation process, so it proposed a new joint transfer network for unsupervised bearing fault diagnosis from the simulation domain to the operational field. The digital twin approach was used by \cite{ert3} in the intelligent manufacturing area to monitor the health of a rotating machine. The main challenge of this study was the nonlinear dynamics and uncertainty involved in the machine degradation process. It successfully overcame this challenge and achieved the desired goal, as evidenced by the results.

Round 2

Reviewer 1 Report

This is an interesting paper that providing a summary of the condition monitoring of Induction Motor and Digital Twin. This paper proposes a predictive maintenance approach based on the establishment of digital twin model of Induction Motor. There are several main contributions in this paper: First of all, this paper reviews current studies for condition monitoring of Induction Motor and Digital Twin. Then the digital twin model for Induction Motor is established with Simulink software. Last but not the least, the proposed diagnostic protocol is clarified and the effectiveness of the method is proven.

The authors have addressed my concerns and issues that I raised. I think the authors should cite more recent papers on predictive maintenance to highlight the contributions, for example, “Optimal selective maintenance scheduling for series-parallel systems based on energy efficiency optimization” (2022); “Fleet-level opportunistic maintenance for large-scale wind farms integrating real-time prognostic updating” (2021); “Long-term predictive opportunistic replacement optimization for a small multi-component system using partial condition monitoring data to date” (2020). So I am glad to recommend this paper to be accepted after minor revision.

Reviewer 2 Report

The comments were satisfactorily addressed, therefore I recommend the manuscript for publication.