You are on page 1of 4

1. Considering a single different parameter cannot be considered as a novel work (i.e.

, tool
wear in place of acoustic emission). There is no novelty found in this work.
T. Tamizharasan, J. K. Barnabas, and V. Pakkirisamy [1] performed optimization in turning
process by using acoustic emission where the parameters considered were material hardness,
cutting speed, feed and depth of cut. Using simulated annealing method they found the
optimum cutting parameters, in addition they also performed the prediction test and compared
with the experimental values. It was further mentioned in the paper that regression analysis
will fail to find the global optimum point. Optimization of turning process using acoustic
emission can be found in a traces in literature also [2], [3].
In this paper a main distinguishing factor is that certain outputs measured from sensors (cutting
force, tool wear, tool vibration) were also used to train the neural network this will allow for a
better prediction of the result. This was referred as sensor fusion techniques and it was used
initially by R. Azouzi and M. Guillot [4] in regression analysis and it was further proposed in
the paper that sensor fusion analysis have led to a better prediction. Here, in this paper the idea
was sensor fusion was infused with artificial neural network to further increase the accuracy of
prediction. Apart from it, the material that has been considered for this work was SS410 steel,
optimization of cutting parameters for the same has not been performed (based on the literature
survey).
2. At present, deep-learning are being considered as a powerful tool to solve such type of
problems. Why the authors are limited to feed-forward back propagation neural network,
which is now considered as a basic tool for prediction.
As stated by the reviewer deep learning is considered as a powerful tool compared to the basic
feed-forward back neural network. But there are many evidences which states that deep
learning increases the computation time and also the increases the complexity [5]–[7].
Furthermore deep learning is used in places where the data set is on a larger magnitude as stated
previously. Here the basic neural network itself is able to predict the answer with higher
accuracy thereby deep learning was not used. Deep learning is used as a tool in places where
the feature sets cannot be defined and the layers present in deep learning takes or makes
decision on the selection of features. Though the term features is used in the application of
image processing it can be converted or the similar terminology can be taken as machining
parameters for this paper.
3. In section - 1, para - 2 - 'In the present study...non-conventional intelligent computational
methods...'. Why non-conventional? Explanations are required.
In this context the word conventional was referred to the statistical methods. It was stated non-
linarites occur machining process and regression method has a comparatively low accuracy.
The comparative study performed in this research work between ANN and statistical regression
method also suggest that regression has a lower accuracy because of the presence of these non-
linarites.
4. In section 2.2, most of the equations are illegible.
The equations have been suitably modified for legibility.
5. In section 2.2., para -3 - what is D(𝜖) and D(𝜌)? They are not explained anywhere in the
manuscript.
𝜖 was assumed to be a sequence of independent random variables with uniform distribution of
probability where the mean equals 0 and the variance is 𝐷2 (𝜖) (for relative displacement of
tool and workpiece)
𝜌 was assumed to be a sequence of independent random variables with uniform distribution of
probability where the mean equals 0 and the variance is 𝐷2 (𝜌) (for instantaneous cutter
position resulting from the face run-out)
Both the above mentioned has been added in the research work.
6. In Figure 1's title - 'Algorithm of steps...'. Algorithm itself consists of steps. It needs
modification. Further, the steps in Fig.1 require further clarification.
The faulty statement present in the current paper has been modified and the correct statement
has been incorporated.
7. In Section 4.2, para -1 - 'Sigmoidal function was used...'. Why sigmoidal function used for
this application? Need further explanations.
It can be found in many literatures that sigmoidal function was used to train the neural network.
The core yoke of using it, was that they enable non-linearity in the network [8]. As it has been
mentioned in the paper that non-linarites are very common in the case of machining sigmoidal
function will be able to cope up compared with other available functions.
8. In section 4.2, para - 3 - 'It has four hidden layers... to surface roughness'. How can you
determine the number of hidden layers in an ANN? Need further explanations.
It was stated by Wang et al. [9] that there are no general rules for specifying the number of
layers in hidden layers. It was stated that the number of hidden layers can be varied from two
to twelve, but as the hidden layers increased training cycle duration was also found to be
increasing [10]. And further it was stated that the influencing parameter is the number of nodes
that are present in the hidden layer and not the number of hidden layers [11]. And in all the
above mentioned literatures it was stated that the number of hidden was determined by trial
and error method, but increasing the number of hidden layer will increase the training time.
Thus hidden layer stated in this paper was also found by trial and error method, by ensuring
that number of the hidden layer is in the minimum level and also is able to predict with better
accuracy.
The required explanation and its corresponding bibliography has been added to the paper.
9. Section 5, para -3 - 'The reason behind...produced during machining'. This sentence does
not make any sense. Require further explanations.
The reason behind …….. produced during machining has been replaced completely for better
understanding and the replaced sentence is as follows:
It has been stated in literature that non-linarites were present during machining. The presence
of non-linarites has affected the accuracy of predicted values in the case of regression, whereas
ANN has the capacity to cope up with the non-linarites.
10. Authors can show the effects on surface roughness prediction by gradually reducing the
number of input parameters to improve the quality of the work.
The work suggested by the reviewer has been carried out in certain research work [4], [9]–[11],
this paper was an attempt to improve accuracy of prediction by fusing the sensor readings into
artificial neural network.
11. The manuscript has several grammatical errors which should be properly corrected before
submission.
The paper has been modified appropriately and all grammatical errors present were corrected.

Reference:
[1] T. Tamizharasan, J. K. Barnabas, and V. Pakkirisamy, ‘Optimization of turning parameters
by using design of experiments and simulated annealing algorithm based on audible
acoustic emission signals’, Proc. Inst. Mech. Eng. Part B J. Eng. Manuf., vol. 226, no. 7,
pp. 1159–1173, Jul. 2012.
[2] W. König, K. Kutzner, and U. Schehl, ‘Tool monitoring of small drills with acoustic
emission’, Int. J. Mach. Tools Manuf., vol. 32, no. 4, pp. 487–493, Aug. 1992.
[3] B. M.s.h., C. I.a., and N. Y., ‘An innovative approach to monitor the chip formation effect
on tool state using acoustic emission in turning’, Int. J. Mach. Tools Manuf., vol. 58, pp.
19–28, Jul. 2012.
[4] R. Azouzi and M. Guillot, ‘On-line prediction of surface finish and dimensional deviation
in turning using neural network based sensor fusion’, Int. J. Mach. Tools Manuf., vol. 37,
no. 9, pp. 1201–1217, Sep. 1997.
[5] P. Maji and R. Mullins, ‘On the Reduction of Computational Complexity of Deep
Convolutional Neural Networks’, Apr. 2018.
[6] N. Nojiri, X. Kong, L. Meng, and H. Shimakawa, ‘Discussion on Machine Learning and
Deep Learning based Makeup Considered Eye Status Recognition for Driver Drowsiness’,
Procedia Comput. Sci., vol. 147, pp. 264–270, Jan. 2019.
[7] C. Ning and F. You, ‘Optimization under uncertainty in the era of big data and deep
learning: When machine learning meets mathematical programming’, Comput. Chem.
Eng., vol. 125, pp. 434–448, Jun. 2019.
[8] K. C. Luk, J. E. Ball, and A. Sharma, ‘A study of optimal model lag and spatial inputs to
artificial neural network for rainfall forecasting’, J. Hydrol., vol. 227, no. 1, pp. 56–65,
Jan. 2000.
[9] X. Wang, W. Wang, Y. Huang, N. Nguyen, and K. Krishnakumar, ‘Design of neural
network-based estimator for tool wear modeling in hard turning’, J. Intell. Manuf., vol. 19,
no. 4, pp. 383–396, Aug. 2008.
[10] U. Zuperl and F. Cus, ‘Optimization of cutting conditions during cutting by using neural
networks’, Robot. Comput.-Integr. Manuf., vol. 19, no. 1, pp. 189–199, Feb. 2003.
[11] F. J. Pontes, A. P. de Paiva, P. P. Balestrassi, J. R. Ferreira, and M. B. da Silva,
‘Optimization of Radial Basis Function neural network employed for prediction of surface
roughness in hard turning process using Taguchi’s orthogonal arrays’, Expert Syst. Appl.,
vol. 39, no. 9, pp. 7776–7787, Jul. 2012.

You might also like