Torque Ripple Suppression of Switched Reluctance Motor with Reference Torque Online Correction
Round 1
Reviewer 1 Report
Dear Authors, please have a look to the explanation requests in the attached file.
Comments for author File: 
Comments.pdf
Author Response
Response to Reviewer 1 Comments
Point 1: Page 4, line 111, Comment, “Current can be calculated out by Eq. (4)” should be changed to “Current can be calculated out by Eq. (5)” .
Response 1: Thank you for careful reading, we have correct the mistake.
Point 2: In page 4. Comment, “Sorry, according to me, you cannot show some results of the machine operation if you have not shown a cross section of the machine and a table with its main geometrical and operational data.”
Response 2: Thank you for your kindly advise, we change the Figure 2. to Figure 5. And the corresponding description is changed to the end of the TRNN structure description. As the results is a machine operation according to Figure 4, which is the proposed TRNN structure.
Point 3: In page 4. Comment, “Sorry, referred to which control scheme? The one of Fig. 3? If so, put Fig. 3 before Fig. 2.”
Response 3: Thank you for your kindly advise, the “Curve of reference torque and instantaneous torque” refer to the Figure 4. TRNN structure, which is labeled as Figure 3 in the original manuscript.
We also revise the description to “The four curves, generating according to the control structure in Figure 4, from top to bottom are reference and instant torque, torque error and rotor angle, respectively.” at Page 6, line 151.
Point 4: In page 5, line 124. High light sentence “the reference torque is hysteresis to the real-time torque.”
Response 4: Sorry, this sentence is not clear, so we change this sentence as “the reference torque lags behind the instant torque.” at page 6, line 155. In the system, the reference torque generated first, then the driving signal can be obtained, followed by the instant torque generating.
Point 5: In page 5. Comment, “It is not clear what are the control schemes of Fig.s 2 and 3. Are they other schemes found in literature? Or are you showing the steps that brought you to the final scheme of Fig. 4? Please, explain this clearly at the beginning of Section 3 ”
Response 5: Sorry, we haven't explained the origin of the PD compensation algorithm clearly. PD compensation algorithm is our comparison algorithm. The author of literature [31] proposed it, and we made a supplementary explanation in the paper as follows. In page 4, line 131.
“Since the final control object is current, the method of transforming torque error to the given current is designed. The instantaneous torque is introduced to compensate for the given current in advance, and the PD controller is used to compensate [31].”
Point 6: In page 6. Comments, “Define RBF ” and “Not clear for a person who does not know NN ”
Response 6: Thank you for your kind advice. We add one of the definitions of RBF and explain how it works as below. In page 7, line 169.
“RBF neural network is a neural network with radial basis function as its core. It is usually designed as a three-layer structure, including an input layer, hidden layer, and output layer. The neuron activation function of the hidden layer is composed of radial basis functions. The array operation of the hidden layer is called the hidden layer node. Each hidden layer contains a central vector c, which has the same dimension as the input vector X. The radial basis function is usually selected as the Gaussian function, which is
h(i)= exp(-r(i)^2/(2*b(i)^2))
where r(i)=||X-c(i)|| is also known as Euclidean distance, c(i) is the center vector, and b(i) is the network width.
The training of the RBF network is the learning process of network parameters, including the center of the hidden layer, network width, number of hidden layer nodes, and the connection weight value from the hidden layer to the output layer. The hidden layer center is generally trained by random selection or unsupervised clustering based on data samples.”
Author Response File: Author Response.pdf
Reviewer 2 Report
The paper is well-written and extensive work has been done. However, please refer to the following points if it helps to improve the paper.
1. Adding an comprehensive summary of results in tabular form.
2. How can the ripple be reduced furhter?
3. Since the simulation is based on MATLAB, it must be checked and verified using either an electromagnetic/FEM model to see the response of the machine.
4. The proposed model performance can be verified with prototype either at this stage or in future.
Author Response
Response to Reviewer 2 Comments
The paper is well-written and extensive work has been done. However, please refer to the following points if it helps to improve the paper.
Point 1: Adding an comprehensive summary of results in tabular form.
Response 1: Thank you for kindly advise. In page 13, a tabular form-- table 2 is added to compare and summary the three method. We also add the summary as
“The fuzzy compensation algorithm can optimize the reference torque, which can combine well with the human experience, but searching the fuzzy rule table consumes more processor time and increases the algorithm's complexity. PD algorithm is a model-free compensation algorithm that realizes the calculation from torque error to compensation current. The computation is small, but its parameters are fixed, so the adjustment amount is fixed. The TRNN algorithm aims at optimizing the reference torque, has the ability of self-learning, needs model information, and requirements for rotor angle accuracy.”
Point 2: How can the ripple be reduced furhter?
Response 2: Thank you for your kindly advise.
The goal of one-step optimization has been achieved based on the proposed reference torque neural network (TRNN). According to the principle of iterative optimization, the proposed neural network can be improved into a Recurrent Neural Network. As the current commutation period of motor operation is an iterative period, the technical optimization of the previous period's data can be achieved. Through iterative optimization, the optimized phase current can be further obtained, and the torque ripple can be reduced.
Point 3: Since the simulation is based on MATLAB, it must be checked and verified using either an electromagnetic/FEM model to see the response of the machine.
Response 3: Thank you for your kindly advise.
During the simulation, we considered the actual motor's response characteristics. The torque table used in the simulation is based on the rotor lock experience, that is, the data obtained from the actual motor test. The instantaneous torque in the simulation is established by testing the actual electromagnetic torque by fixing the rotor at different rotor angles and providing different currents. The electromagnetic torque in simulation is also provided by the established torque table. It corresponds to the real-world motor output. At present, due to the breakdown of the drive circuit, and although we try our best to repair the drive circuit, we still can not show the phase current and electromagnetic torque curve in operation.
We also considered the algorithm's complexity and whether it can run on the embedded platform. The answer is YES. The neural network algorithm we proposed has a small number of nodes, a small amount of calculation when updating the weight, and a low calculation requirement for the processor when it is used in the actual control platform, so its calculation speed can be relatively close to the situation of the simulation platform.
Point 4: The proposed model performance can be verified with prototype either at this stage or in future.
Response 4: Thank you for your kindly advise. During the simulation, we did consider the problem of prototype verification. However, the experimental platform is broken at present. Therefore, we use the data collected formerly based on the platform for simulation verification.
Next, we will continue to improve the algorithm and improve performance of the algorithm. Of course, we will also repair the driver circuit of the platform in order to complete the subsequent algorithm verification on the prototype.
Author Response File: 
Author Response.docx
Reviewer 3 Report
The manuscript proposes a reference torque neural network (TRNN) with reference torque online correction for suppressing torque ripple of the switched reluctance motor (SRM). The proposed TRNN is a single input and output network based on the torque sharing function (TSF) method. The proposed method is validated by MATLAB simulations.
The manuscript is well-written and easy to follow. Moreover, it is graphically well-equipped.
Here are some comments I would like the authors to address before the manuscript is considered for publication:
1. The correct rotor angle information is essential in the proposed approach as it is used as input to the method. Therefore, the authors should extend the literature review in the Introduction section by mentioning recent studies on determining rotor angle in salient synchronous machines. Please consider briefly mentioning the following papers: 10.1109/TPEL.2022.3162963; 10.3390/en12091609; 10.3390/electronics9091371.
2. The proposed method is validated only by numerical simulations in MATLAB/Simulink. However, this type of method can often be genuinely validated by experimental tests on the real-world drive. Therefore, the authors should provide some experimental validation of the proposed method.
3. The performance of the proposed method in the simulations is compared only to one alternative approach – PD compensation. However, the authors should compare the proposed method to more state-of-the-art techniques, analyze results and provide a discussion.
4. Additionally, the performance of the proposed method should be analyzed at more operating points (speed and torque), not just two. This additional analysis is required to support the effectiveness of the proposed method and allow more general conclusions.
5. In the Conclusion section, the authors should elaborate on some limitations of the presented study.
6. In the Conclusion section, the authors should also provide some directions for future research.
Author Response
Response to Reviewer 3 Comments
The manuscript proposes a reference torque neural network (TRNN) with reference torque online correction for suppressing torque ripple of the switched reluctance motor (SRM). The proposed TRNN is a single input and output network based on the torque sharing function (TSF) method. The proposed method is validated by MATLAB simulations.
The manuscript is well-written and easy to follow. Moreover, it is graphically well-equipped.
Here are some comments I would like the authors to address before the manuscript is considered for publication:
Point 1: The correct rotor angle information is essential in the proposed approach as it is used as input to the method. Therefore, the authors should extend the literature review in the Introduction section by mentioning recent studies on determining rotor angle in salient synchronous machines. Please consider briefly mentioning the following papers: 10.1109/TPEL.2022.3162963; 10.3390/en12091609; 10.3390/electronics9091371.
Response 1: Thank you for your recommendation, the three papers can rich the literature review and they are cited as reference 15, 16 and 34, respectively.
We quote as
“However, this method needs to adapt better to load changes and the accuracy of rotor angle is required [15]. Therefore, the accuracy of rotor angle needs to be considered [16].”
“The robustness of the system to speed control can be enhanced by introducing a neural network into the sensor-less speed [34].”
Point 2: The proposed method is validated only by numerical simulations in MATLAB/Simulink. However, this type of method can often be genuinely validated by experimental tests on the real-world drive. Therefore, the authors should provide some experimental validation of the proposed method.
Response 2: Thank you for your kindly advise.
During the simulation, we considered the actual motor's response characteristics. The torque table used in the simulation is based on the rotor lock experience, that is, the data obtained from the actual motor test. The instantaneous torque in the simulation is established by testing the actual electromagnetic torque by fixing the rotor at different rotor angles and providing different currents. The electromagnetic torque in simulation is also provided by the established torque table. It corresponds to the real-world motor output. At present, due to the breakdown of the drive circuit, and although we try our best to repair the drive circuit, we still can not show the phase current and electromagnetic torque curve in operation.
We also considered the algorithm's complexity and whether it can run on the embedded platform. The answer is YES. The neural network algorithm we proposed has a small number of nodes, a small amount of calculation when updating the weight, and a low calculation requirement for the processor when it is used in the actual control platform, so its calculation speed can be relatively close to the situation of the simulation platform.
Point 3: The performance of the proposed method in the simulations is compared only to one alternative approach – PD compensation. However, the authors should compare the proposed method to more state-of-the-art techniques, analyze results and provide a discussion.
Response 3: Thank you for your kindly advise, we searched the relevant literature, added a fuzzy compensation algorithm proposed in literature 32, which is related to our algorithm, and we compare it with our proposed method. At the same time, we explained the source of the PD compensation algorithm in more detail.
It should be noted that, the method proposed in this paper is online neural network compensation. There needs to be more literature on online learning and compensation based on neural networks. Most of the methods based on neural network optimization are offline optimization. However, offline optimization requires many data and considering the data migration problem, the data trained on one motor may not possible migrate to another. Therefore, the online and offline methods must be more than just compared to the performance.
Point 4: Additionally, the performance of the proposed method should be analyzed at more operating points (speed and torque), not just two. This additional analysis is required to support the effectiveness of the proposed method and allow more general conclusions.
Response 4: Thank you for your suggestion. We have done more tests under different speeds and torque loads, and the torque ripple comparison is shown in Table 1 (page 13) and Figure 12 (page 15) . All tests show that the proposed TRNN algorithm effectively reduces torque ripple. However, it should be noted that the current hysteresis control method is generally used at medium and low speeds, so the speed we tested is up to 900 rpm.
Point 5: In the Conclusion section, the authors should elaborate on some limitations of the presented study.
Response 5: Thank you for your careful reading. The algorithm proposed in this paper does have shortcomings. What is obvious is that the algorithm is to adjust the reference torque online, which requires high-angle measurement accuracy. We added a description on page 16, line 309.
Point 6: In the Conclusion section, the authors should also provide some directions for future research.
Response 6: Thank you for your suggestion. The TRNN algorithm proposed in this paper can optimize the reference torque in one step, requiring high-angle measurement accuracy. In future research, TRNN can be improved into a recurrent neural network in which the optimized value of the previous cycle can be used to optimize the next motorcycle. Therefore, online iterative optimization can be realized, and the effect will be further improved.
We added the description on page 16, line 310.
Round 2
Reviewer 3 Report
The authors have addressed my comments.
