Mill Feed Control System and Algorithm Based on Python

Round 1

Reviewer 1 Report

The manuscript “Mill Feed Control System and Algorithm based on Python” contains an interesting case of the application of a Python-based fuzzy controller to the mill feeding process.

I have a few comments regarding the manuscript under consideration:

1.      The literature review should be extended to include items on the use of fuzzy logic in modeling the grinding process.

2.      There is no description of the axis in the following Figures: 5-7, 11, 14, 15.

3.      If possible, an experimental verification of the model should be performed.

Author Response

Dear Reviewer:

Thank you for your comments concerning our manuscript entitled “Mill Feed Control System and Algorithm based on Python”. Those comments are all valuable and very helpful for revising and improving our paper, and provide the important guiding significance to our studies. We have studied the comments carefully and have made corrections, which we hope to meet with approval. We submitted a revised manuscript (Changes highlighted). Those revised portions are marked in red in the paper. The main corrections in the paper and our reply to your comments are listed as following:

Point 1:  The literature review should be extended to include items on the use of fuzzy logic in modeling the grinding process.

Response 1: Thank you very much for this valuable suggestion. We edited the introduction according to your suggestion.The literature on fuzzy logic modeling of grinding process has been supplemented.

Point 2: There is no description of the axis in the following Figures: 5-7, 11, 14, 15.

Response 2: Thank you very much for this valuable suggestion.The uncommented diagram has been modified.

Point 3: If possible, an experimental verification of the model should be performed.

Response 3: Thank you very much for this valuable suggestion.Modeling and simulation should be carried out in the early stage, and the field application should be considered in the later stage, and corresponding improvement should be made.

We appreciate reviewers’ warm work earnestly, and hope that the corrections will meet with your approval.

Once again, thank you very much for your comments and suggestions.

Author Response File: Author Response.pdf

Reviewer 2 Report

Overall comments

The authors report a fuzzy control algorithm for mill feeding based on Python.

A Kalman filter is used to filter and correct the estimation error of weight measurements made by an electronic belt scale. A linear regression is used to identify the most suitable input variables, and the resulting control system is comprised of two inputs (ore particle size and electronic belt scale weight) and one output (motor frequency of the electronic belt). A fuzzy controller is developed and compared to traditional PID control.

The proposed case studies for comparison are:

- compare the step response of both methods

- compare the responses to change in the field production conditions of both methods

The work addresses a relevant problem in the mining industry, namely the control of mill feeding circuits. The authors consider the weight on the feeding belt and ore particle size as manipulated variables and the motor frequency of the feeding belt as controlled variable. The control goal is to ‘dynamically balance the belt control’. A fuzzy controller is proposed and compared to traditional PID control. The authors use Python to model and implement the controller.

I believe that this work can be improved before being accepted for publication, and thus reach and contribute to the research of many more readers once it has been reworked. The authors can find a few comments to guide them through the review process below.

Thanks for your hard work and I hope to see your paper accepted soon.

Comments by section

Introduction

- ‘and most of them are associated, and hence ore tends to be lean and difficult to dress.’ I don’t understand what the authors mean by associated ore deposits and ore is difficult to dress.

- ‘and has greatly promoted the research and development of intelligent control algorithms[11].’ I don’t see how reference 11 is related to fuzzy control and development of intelligent control algorithms as reference 11 reports the result of a survey on how ease of use can impact usage of internet banking services.

Error factors in the mill feeding process

- Was Equation 2-2 obtained from the literature or did the authors develop it themselves?

- ‘which was calculated by substituting T into Equation (2-2).’ I think the authors mean that T_e was calculated using Eq. 2-3 and then used as T in Eq. 2-2. Otherwise I don’t understand this sentence.

- ‘Here, p is the output power of the motor, \mu is the friction coefficient of the belt and roller contact surface, \nu is the belt transport speed, and Q is the belt weight per unit length.’ The symbols are not aligned.

- Matrices/vectors ‘weight’ and Q_r in Eq. 2-13 have not been defined.

- ‘and the results are shown in Figures 2.3 and 2.4.’ Figure references are confusing since in some sentences the authors carry the section number and in other sentences they do not.

- ‘the system parameters A and B and measurement parameter H are all 1.’ What are A, B and H?

- ‘the error between the predicted and true values is no more than 1.5%, and the accuracy continues to increase.’ How is the accuracy related to the error between observed and true values, and how can you claim that the accuracy continues to increase if the estimation has stabilized?

Python-based fuzzy control algorithm

- ‘according to the experience of field operators and related literature on beneficiation, the ore particle size, electronic belt scale weight, and motor frequency are considered individually’. It would be interesting to further detail how the authors translated the experience of field operators into the fuzzy rules, as well as which references from the literature have been used.

- Figure 11, missing legend, axis names

- ‘Figure 13 compares the step-response curves of the traditional PID control algorithm and the fuzzy control algorithm’ The simulation scenario is not clear for me. Did the authors simulate a step response in the reference, or are the authors comparing how the controller can stabilize the system after a manual step change in the input? What is the y-axis is Figure 13? 

- in Section 2.3, the sampling times of humidity and belt scale weight were reported as 30 and 10 min, respectively. In Figure 13, the result is reported in a tame scale of seconds. How does the sampling time of the real process affect the performance of the proposed controller?

- is the PID implemented in the simulations also benefitting from error correction in weight measurements from the Kalman filter proposed by the authors?

- PID control is usually used for single input single output processes, but the process chosen by the authors is a multiple input single output process. It would be interesting to have more details about how the PID was implemented and tuned.

- ‘Figure 3.9 shows that the overshoot of the traditional PID control’ which is Figure 3.9?

- for the analysis in Figure 16, it would be interesting to know if (and at what time) the system is being affected by disturbances, or is the oscillation in overshoot (%) only caused by the measurement error of the weight?

Conclusion

- ‘the simulation results can be displayed in three dimensions, which is more intuitive and convenient for operators to observe and record.’ What sources do the authors use to support this claim?

References

- ‘D. Kufoalor, G. Frison, L. Imsland, T.A. Johansen, J.B.J.J.o.P.C. J?Rgensen, Block factorization of step response model predictive control problems, 53 (2017) 1-14.’, problem printing characters.

Author Response

Dear Reviewer:

Thank you for your comments concerning our manuscript entitled “Mill Feed Control System and Algorithm based on Python”. Those comments are all valuable and very helpful for revising and improving our paper, and provide the important guiding significance to our studies. We have studied the comments carefully and have made corrections, which we hope to meet with approval. We submitted a revised manuscript (Changes highlighted). Those revised portions are marked in red in the paper. The main corrections in the paper and our reply to your comments are listed as following:

Comments by section:

1、Introduction

Point 1: ‘and most of them are associated, and hence ore tends to be lean and difficult to dress.’ I don’t understand what the authors mean by associated ore deposits and ore is difficult to dress.

Response 1: Thank you very much for this valuable suggestion.

What is associated mineral: Associated mineral is a mineral deposit containing other minerals;  Generally, mineral deposits contain associated minerals, but if the associated content is generally not too high, only in the case of large value mining separation. 

What are associated deposits: Associated deposits are minerals that are more readily available.  They are more common in the ore deposit distribution, local enrichment of also can close to the extent of the mineral industry index requirements, in the processing of main mineral beneficiation process can choose them at the same time the product or intermediate product qualified concentrate: some important mineral content in the ore is low, but because of the special needs or a high economic value will also be necessary to recycle.  Since concentrate products can be separated, the recovery and utilization of ore components can be enlarged, which is of great significance to further improve the economic value of the deposit.

Single deposits, which are easy to mine and sort, are relatively rare.

Point 2: ‘and has greatly promoted the research and development of intelligent control algorithms[11].’ I don’t see how reference 11 is related to fuzzy control and development of intelligent control algorithms as reference 11 reports the result of a survey on how ease of use can impact usage of internet banking services.

Response 2: Your suggestion is important. The reference has been reread and removed.

• Error factors in the mill feeding process

Point 1: Was Equation 2-2 obtained from the literature or did the authors develop it themselves?

Response 1: Thank you very much for this valuable suggestion.This formula is obtained from the literature,“A new method which can offset the tension of belt and improve the measuring precision of belt scale”

Point 2: ‘which was calculated by substituting T into Equation (2-2).’ I think the authors mean that T_e was calculated using Eq. 2-3 and then used as T in Eq. 2-2. Otherwise I don’t understand this sentence.

Response 2: Your suggestion is important.T in 2-2 and Te in 2-3 are actually a T, which can be calculated by formula 2-3 to get T. By substituting T into formula 2-2, F can be obtained.Formulas 2-3 have been modified.

Point 3: Here, p is the output power of the motor, \mu is the friction coefficient of the belt and roller contact surface, \nu is the belt transport speed, and Q is the belt weight per unit length.’ The symbols are not aligned.

Response 3: Thank you very much for this valuable suggestion. Modify the input method. The input method has been modified in the corresponding position.

Point 4: Matrices/vectors ‘weight’ and Q_r in Eq. 2-13 have not been defined.

Response 4: Your suggestion is important.Q_ weight is the weight covariance of the sensing measurement, Q_ r is the belt slip covariance.We have compensated for this problem in the article.

Point 5: ‘and the results are shown in Figures 2.3 and 2.4.’ Figure references are confusing since in some sentences the authors carry the section number and in other sentences they do not.

Response 5: Thank you very much for this valuable suggestion.This problem occurred because it was missed when modifying the format, which has been corrected.

Point 6:  ‘the system parameters A and B and measurement parameter H are all 1.’ What are A, B and H?

Response 6: Your suggestion is important.

Suppose you plant a fruit tree when you are traveling to a scenic spot, but it is obvious that the process from seedling to fruit can not be completed in a day or two. It needs to grow slowly. And you can't go to see it often, but you want to know the height of the fruit trees. What should you do?

So the problem to be solved is: how to correctly estimate the height of a fruit tree?

We call the height of the fruit tree we want to know as the state variable to be estimated. We want to know once a year, which is called step size. You need to know the approximate height when planting fruit trees, so you think about it, about 1 meter, but I'm not sure. It may be 90 cm or 110 cm. So this 1 meter is called your initial state estimation, and this 10 cm uncertainty is called the error covariance matrix of the state estimation, which will change with your next estimation.

What shall I do?

You have checked on the Internet. This kind of fruit tree grows 10% higher than the previous year almost every year (pure assumption). This growth law / model is called the state transition matrix. So according to the model, if there is one meter in 2017, there will be about 1.1 meters in 2018. But obviously, this model is not applicable to any fruit tree and the actual sunshine, wind and soil conditions in your local area, and you know that fruit trees cannot grow indefinitely. So our model can't be 100% accurate. We use -- a thing called process noise to measure it. It can be understood as the inaccuracy of model recurrence. For example, suppose that the error of this estimation model is 0.3m, which is called white noise. The smaller the process noise is, the more you believe that the growth model is accurate. The larger the process noise is, the more garbage the model.

Point 7: ‘the error between the predicted and true values is no more than 1.5%, and the accuracy continues to increase.’ How is the accuracy related to the error between observed and true values, and how can you claim that the accuracy continues to increase if the estimation has stabilized?

Response 7: Thank you very much for this valuable suggestion.The real value is the set value of the system, so we need the system to reach this value, and there are two values in the Kalman filter: one is the observation value of the sensor, and one is the predicted value of the Kalman filter.  Depending on the Kalman gain you have a choice between believing the observed value more or believing the predicted value more.

Based on the state space representation of linear system, the optimal estimation of system state is obtained from the output and input observation data.  The system state mentioned here is a set of minimum parameters that summarizes the effects of all the past inputs and disturbances on the system. Knowing the system state, we can determine the whole behavior of the system together with future inputs and disturbances of the system.

3、Python-based fuzzy control algorithm

Point 1: ‘according to the experience of field operators and related literature on beneficiation, the ore particle size, electronic belt scale weight, and motor frequency are considered individually’. It would be interesting to further detail how the authors translated the experience of field operators into the fuzzy rules, as well as which references from the literature have been used.

Response 1: Thank you very much for this valuable suggestion.The article has been supplemented accordingly.

Point 2: Figure 11, missing legend, axis names.

Response 2:Your suggestion is important.Figure 11 has been modified and supplemented accordingly.

Point 3:‘Figure 13 compares the step-response curves of the traditional PID control algorithm and the fuzzy control algorithm’ The simulation scenario is not clear for me. Did the authors simulate a step response in the reference, or are the authors comparing how the controller can stabilize the system after a manual step change in the input? What is the y-axis is Figure 13?

Response 3: Thank you very much for this valuable suggestion.We are looking at the response time and amplitude of the two systems from 0 to 1 to see when they will stabilize. It is equivalent to doing the step response simulation experiment first to see if it is effective under the step response state, and then doing the simulation experiment.

The y-axis in Figure 13 is the amplitude.

Point 4: in Section 2.3, the sampling times of humidity and belt scale weight were reported as 30 and 10 min, respectively. In Figure 13, the result is reported in a tame scale of seconds. How does the sampling time of the real process affect the performance of the proposed controller?

Response 4: Thank you very much for this valuable suggestion.The sampling time is set to 30 minutes to collect data. Collect data and input it into the regression model to see the size of the influencing factors, which is equivalent to the collection time set to know the size of the influencing factors.

In Figure 13, the second unit is the algorithm requirement of step response, which is not related to the collected data in Section 2.3.

The purpose of data acquisition is to determine the input and output values of the fuzzy controller, not to make a step response. The step response has its own independent algorithm and does not need to collect field data for input. It only needs to compare the algorithms of the two comparison models.

Point 5: is the PID implemented in the simulations also benefitting from error correction in weight measurements from the Kalman filter proposed by the authors?

Response 5: Your suggestion is important.We first process the collected data with Kalman filter, and then give the data to PID control and fuzzy control, so Kalman filter is also beneficial to PID control.

Point 6: PID control is usually used for single input single output processes, but the process chosen by the authors is a multiple input single output process. It would be interesting to have more details about how the PID was implemented and tuned.

Response 6: Thank you very much for this valuable suggestion.We are currently studying this issue, which is expected to be reflected in the next article.

Point 7: ‘Figure 3.9 shows that the overshoot of the traditional PID control’ which is Figure 3.9?

Response 7: Your suggestion is important.This is a typographical error in our format, which has been corrected in the article.

Point 8: for the analysis in Figure 16, it would be interesting to know if (and at what time) the system is being affected by disturbances, or is the oscillation in overshoot (%) only caused by the measurement error of the weight?

Response 8: Thank you very much for this valuable suggestion. Not only the weight factor, but also other factors, such as ore sticking on the belt scale and ore feeding gate, will have an impact. Here, only the weight factor has a great influence, and then Kalman filter is used to filter it out. There will still be other factors.

4、Conclusion

Point 1: ‘the simulation results can be displayed in three dimensions, which is more intuitive and convenient for operators to observe and record.’ What sources do the authors use to support this claim?

Response 1: Thank you very much for this valuable suggestion. 

The three-dimensional display is more intuitive to see the interaction between two when the three conditions affect each other. If the two-dimensional image is used, multiple images are required. When viewing multiple images, it is easy to cause interference, and wrong or missed views may occur. Therefore, compared with the two-dimensional graph, the three-dimensional graph is more convenient and intuitive in recording data.

• References

Point 1: ‘D. Kufoalor, G. Frison, L. Imsland, T.A. Johansen, J.B.J.J.o.P.C. J?Rgensen, Block factorization of step response model predictive control problems, 53 (2017) 1-14.’, problem printing characters.

Response 1: Thank you very much for this valuable suggestion. For this problem, corresponding modifications have been made in the article.

We appreciate reviewers’ warm work earnestly, and hope that the corrections will meet with your approval.

Once again, thank you very much for your comments and suggestions.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Dear authors, thanks for working hard and discussing my comments. I believe that your revised manuscript is much clearer now. In particular, I like how you connected the fuzzy rules to the practical operation strategies in the mining industry.

I'm glad that you have ideas for  future work and I hope to read more from you soon.

I'm recommending acceptance in the present form, but please consider the following if you have the opportunity to make changes before submitting the proofs to the journal.

Error factors in the Mill feeding process

Point 6, Sorry for not being clear. I meant that you did not explain what A, B and H are in your manuscript before using them.

I believe that A and B are the dynamic matrices of the state-space function 2-10 and H was defined earlier as 'H is the lifting height of the ore material'.

Conclusions

Point 1, I'm wondering if the plant operators actually consider that a 3d figure is easier to interpret and more useful to implement a control strategy. It would be interesting if you have any information about this.