Post-processing for Three Class of Tool Wear Prognosis using Two

Class ANN Classifier based on Vibration of CNC Milling

Anis Arendra

,Sabarudin Akhmad

, Herianto

, and Kukuh Winarso

Mechanical Engineering Departement, University of Trunojoyo Madura, Indonesia

Industrial Engineering Departement, University of Trunojoyo Madura, Indonesia

Deptement of Mechanical and Industrial Engineering, Universitas Gadjah Mada, Indonesia

Keywords: Tool Wear Prognosis,MultiLayer Perceptron, Vibration, Multilevel Classifier.

Abstract: This research propose a novel method of utilizing bi-levels tool wear classifiers to prognose three levels of

tool wear through additional post-processing stages. The classifier uses a multi-layer perceptron (MLP),

single hidden layer, trained using the resilient backpropagation method. The original classifier output range

-1 to 1 and threshold 0.0 for the separator of two classes, has been able to achieve 100% classification

accuracy of two CNC tool conditions, severe wear and normal one, based on vibration features in the time

domain and order domain. This classifier was tried to classify three levels of tool wear: normal, moderate

wear, severe wear, according to ISO 8688 standard. Output of existing MLP classifier is passed through a

moving average filter with period 4 and using threshold of -0.8 and +0.8 for three level separation, normal

tool, moderate wear, severe wear. The proposed method is proven to achieve 89.98% accuracy from 459

tests. Fail safe missclassification occurred from 153 test cases which were supposed to be moderate wear,

46 of them were incorrectly indicated as severe wear. For the severe wear test case and normal tool test

case, no prediction errors were found. The 100% accuracy for both test case prediction.

1 INTRODUCTION

The CNC machining role has wide spread and

increasingly important in the manufacturing industry

(Cheng et al., 2019) and (Teti et al., 2010), as well

as tool condition monitoring technology (Huang et

al., 2019) and (Zhou et al., 2019). Tool Condition

Monitoring (TCM) technology is constantly being

developed to ensure the quality and efficiency of

machining(Ahmad and Kamaruddin, 2012). The

TCM method is generally divided into: qualitative-

based method, model-based method, data-based

method(Cheng et al., 2019). Model-based methods

analytically build the mathematical models to

explain the phenomenon of machining tool wear,

like the method used by Mishra, (2015), Rmili et al.,

(2016), Liu et al., (2010), and Mei et al., (2018). But

this is not an easy build, given the complexity of the

mechanical machining system(Yau et al., 2014) and

(Huang et al., 2019). While the data-based method

does not require analytical knowledge, it only

requires empirical knowledge about the relationship

of tool wear with physical phenomenon of the

machining (Jemielniak et al., 2012), so this method

is more practical to use (Wei and Wang, 2019).

2 PREVIOUS WORK

Sensors commonly used in tool wear detection are

acoustic emission sensors and accelerometer(Murat

et al., 2017). Lembke, (2019) and Casoli,

(2019)sated, data-based methods of TCM require

classifiers that are supervisically trained as used in

research by Zhou et al., (2019), Casoli et al., (2019)

and Pappachan et al., (2017), or unsupervisically-

trained based on databases as used in research by

Barraza, (2017), Sakthivel et al., (2014) and

Benkedjouh et al., (2017). Commonly used

classifiers are Support Vector Machine (SVM) as

used by Zhou et al., (2019), Artificial Neural

Network (ANN) used by Arendra and Herianto,

(2020), Arendra et al., (2020), and Prasetyo et al.,

(2018), K-Nearest Neighborhood (KNN) used by

Junior et al., (2018), Genetic Algorithm used by Goti

et al., (2019), and Bayesian Network used by Tobon-

Mejia et al., (2012). The supervised training method

Arendra, A., Akhmad, S., Herianto, . and Winarso, K.

Post-Processing for Three Class of Tool Wear Prognosis using Two Class ANN Classiﬁer based on Vibration of CNC Milling.

DOI: 10.5220/0010307500003051

In Proceedings of the International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies (CESIT 2020), pages 269-276

ISBN: 978-989-758-501-2

269

for classifier requires a database for training,

validation and testing(Leturiondo et al., 2016). The

number of database subsets for this classifier is the

same as the number of classes to be

grouped(Wiharto et al., 2016).

Multiclass SVM classifier has been used by

Wiharto et al., (2015) and Cheng et al., (2019)for the

classification of five levels of heart disease cases

and the classification of three levels of of CNC tool

wear, respectively. Wiharto et al., (2015)used five

subsets of data at healthy, sick-low, sick-medium,

sick-high, sick-serious levels. Cheng et al.,

(2019)use three subset data at healthy state level,

degradation state level, and failure state level. As

expected, for n class classifiers a number of n

training data sub-sets are required. Yiqian He et al.,

(2019)uses a different approach, exploring the

correlation between the mahalanobis distance of

vector features to the wear level, then setting two

thresholds to separate the three wear levels of tool.

The approach taken by Tobon-Mejia et al., (2012)is

constructing the model behavior of tool degradation

phenomena with a set of mathematical models for

predicting the evolution of tool degradation. This

paper proposes a new approach to classifying three

levels of tool wear using a classifier that is trained

with twower classes, then followed by post-

processing.

3 EXPERIMENT SETUP AND

METHODS

The sensors used to collect data are MEMS

Accelerometer MMA7361 and DT-SENSE Tracking

SFH-300. MMA7361 sensor sensitivity is 800mV /

g in the range of 0g - 1.5g. The sensor output is

analog voltage 0 - 3.3V for X, Y, and Z channels.

This sensor is installed in the workpiece fixture. The

DT-SENSE Line Tracking Sensor SFH-300 is used

as an optical proximity sensor to detect the spindle

rotation phase that has been marked with a white

reflector strip. This sensor works on a 5VDC voltage

power supply, has four channel voltage outputs from

0 to 4.9V, the rise and fall response time is 10µs.

Data acquisition equipment uses DAQ NI USB-

6008 with NI-DAQmx 9.9 device driver. The DAQ

works on USB interface, input voltage range of ±

10V, the maximum aggregate sampling rate for

multichannel is 10kS/s. In this study, DAQ NI USB-

6008 is set in differential analog input mode with a

resolution of 12bits and a maximum of 4 input

channels. Channels 1,2,3, respectively measure the

vibration of the milling machine table in the X, Y, Z

axis. Channel 4 for detection of spindle rotation

phase.

Table 1:Micro photo of end mill tool wear gradation.

Tool Face Major Flank Minor Flank

Norm

Wear

Mode

rate

Wear

Sever

Wear

The experimental treatment used a 4SE-

LIST6210 10X25X75X10 HSS-Co RA26 nachi end-

mill, a four flute end mill tool with a diameter of

10mm. This tool is used for up-milling machining of

mild steel with depth of cut 0.2 mm and 5 mm

cutting width, without coolant. Parameters of

machining spindle speed vary from 550rpm,

650rpm, 750rpm, and feed per tooth varies from

0.02mm, 0.05mm, 0.08mm. The first row and third

row of table 1 show the appearance of normal toll

and severe wear that were used for classifier training

in this study. While the second row of table 1 shows

the appearance of moderate tool wear, for testing

three level classifier alongside with normal tool and

severe wear tool.

4 RESULT AND DISCUSSION

Data acquisition of four channels with a sampling

size of 1024 datapoints was carried out at a sampling

frequency of 2.5 kHz for 409.6 ms. The four

channels consist of X Y Z axis vibrations and

spindle rpm. Feature extraction is performed to get

features represent 3 × 1024 time-domain datasets

and 3 × 513 order-domain datasets. The time-

domain vibration feature is represented by statistical

measure of data distribution, std, skewness, kurtosis,

range for each XYZ axis. Whereas the order-domain

feature is represented by the magnitudes of

acceleration in 1st order to 90th order. Feature

selection is based on correlation analysis, ten

CESIT 2020 - International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies

270

features are selected: stdz, rangey, stdy, rangex,

stdy, rangez, X2nd, Y13th, Z13th, Y2nd.

Figure 1: Discrete waveforms of 1024 time-domain

datapoints, 409.6ms, represented by a statistical measure

of data distribution for the extraction of time-domain

features.

Figure 2: The vibration order spectrum as a result of Fast

Fourier Transform and normalization to the spindle rpm,

for the use of order-domain features extraction.

4.1 MLP Training using Two Level of

Tool Wear

The artificial neural network used for the classifier is

MultiLayerPreceptron (MLP) with a single hidden

layer. In the context of MLP learning and

discovering the general pattern of training data, the

MLP training process is stopped when the training

process begins to show symptoms of overfitting. In

this study, the symptoms of overfitting were

monitored from the MSE of validation sub-set data.

The database of vibration features is divided into

three parts; training data subset to train the classifier,

test data subset to assess the training progress and

validation data subset for early stopping use. During

in the training progress, if the MSE training data

subset decreases but the MSE validation data subset

does not goes down and occurs in 4 epochs

consecutively, then this shows the MLP begins to

learn the specific characteristics of the training data

subset and ignores the generality of the whole data

pattern so that the training process must be stopped.

This MLP training is carried out by setting the

max_fail training parameter by 4 and separate the

dataset in three groups randomly at8: 1: 1 proportion

for the training data subset, the test data subset and

the validation data subset.

A summary of training, testing and validation of

ten MLP is tabulated in Table 2. A performance

comparison of the ten MLP iteration is shown in

Figure 9, in the MSE order of the largest to the

smallest. MLP training with resilient

backpropagation can achieve MSE in the range of

0.0524 to 0.0376. None of the test accuracy in the

group reached 100% accuracy, the highest accuracy

that can be achieved is 97.2%. But the validation

results show the opposite, 3 out of 10 iterations are

able to achieve 100% accuracy. The lowest MSE

training data in the 2nd iteration provides the worst

accuracy in this test group validation. Validation of

100% accuracy is achieved by MLP with MSE

ranging from 0.0524 to 0.0411. The best MLP case

chosen for the classifier is the result of 4th iteration

neuron weighting. The TCM system used in this

study is able to perform the tool wear detection

based on vibration measurements of the CNC

machine.

The detailed output of TCM prediction for severe

tool wear detection with an output target at value of

positive one, is shown in Figure 3. While Figure 5

displays the detailed output of TCM prediction for

normal tool with the target output at value of

negative one. Both images displays 20 predicted

conditions of tool wear for 9 treatment machining

parameter. The first row, second row, third row of

the plot data consecutively are the spindle speed of

750 rpm, 650 rpm, 550 rpm treatment. The first

column, the second column, the third column of the

data plot consecutively are cutting depth parameter

of 0.08; 0.05; 0.02 mm/tooth.

Generally speaking, the MLP predictions have

reached the target, especially on parameters

machining of spindle speeds at 750 rpm and 650

rpm. In the parameters machining of spindle speed at

550 rpm, there are some MLP predictions that are

less close to the target even though they are included

in the right classification. This incidents occurred in

the machining parameter of 0.08 mm/tooth cutting

thickness of severe tool wear condition. MLP

prediction uses a threshold value of 0, then the

positive numbers output of MLP will be concluded

that the tool condition is severe wear, and vice versa.

The negative numbers output of MLP will be

concluded as a normal tool condition. MLP accuracy

for the validation of these 2 classes attain 100%

accuracy for the 360 test cases.

Post-Processing for Three Class of Tool Wear Prognosis using Two Class ANN Classiﬁer based on Vibration of CNC Milling

271

Figure 3: TCM prediction output before post-processing

on the use of severe tool wear for nine machining

parameter.

Figure 4: TCM prediction output before post-processing

on the use of moderate tool wear for nine machining

parameter.

Figure 5: TCM prediction output before post-processing

on the use of normal tool for nine machining parameter.

Figure 6: TCM prediction output with post-processing on

the use of severe tool wear for nine machining parameter.

Figure 7: TCM prediction output with post-processing on

the use of moderate tool wear for nine machining

parameter.

Figure 8: TCM prediction output with post-processing on

the use of normal tool for nine machining parameter.

CESIT 2020 - International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies

272

4.2 Post-processing for Three Level

Classification

The range for deducing tool conditions in the existing

classifier is very wide, above the threshold value of 0

to 1, it is concluded that the condition is severe wear,

and below the threshold value of 0 to -1, it is

concluded that the tool condition is normal. If the

inference output range is narrowed, the inference

range for severe wear condition is set at the limit of

0.8 to 1, and the normal tool condition inference range

is set at the range -1 to -0.8. Then there is a range that

cannot be defined as a severe wear condition or

normal tool conditions, i.e. range above -0.8 to below

0.8. The range between -0.8 to 0.8 is what will be

used as thresholds to the prognosis of a moderate tool

wear.

Figure 9: Training phase MSE, testing phase accuracy and

validation phase accuracy of MLP classifier.

For the validation of three levels classification,

three levels of tool wear were used; normal tool,

moderate wear tool and severe wear tool. The first

tool was a severe wear tool shown in third row of

table 1. The flank wear of this tool has reached the

tool wear criterion based on ISO 8688

recommendations. The second tool is a moderate

worn tool shown in second row of Table 1. The third

tool is a normal tool, there is no significant flank wear

on the cutting edge shown in first row of Table 1.

Detailed output of MLP predictions on the use of

moderate wear tools is shown in Figure 4. MLP

predictions on the use of moderate wear tool show

floating values between -1 to 1, especially in the

machining parameters 0.02 mm/tooth feed and 550

rpm spindle speed. For machining parameters of 0.05

mm/tooth feed, 750 rpm and 0.08 mm/tooth feed,

650 rpm, MLP output tends to scatter in the range of

0 to 1. For machining parameters of 0.08 mm/tooth

feed, 750 rpm , MLP output tends to close to 1.

To reduce the fluctuation of MLP output values

and improve the quality of MLP inferencing, post-

processing is applied by smoothing filter using a

simple moving average with a period of 4. Detailed

MLP conclusion output by smoothing on the use of

severe wear tool, moderate wear tool and normal tool,

consecutively is shown in Figure 6, Figure 7, and

Figure 8. With the application of smoothing post-

processing, MLP output fluctuations are considerably

damped. The MLP prediction output for the use of

severe wear tool is always in the range of 0.8 to 1 for

153 test cases on 9 parameters of machining. With

these results, the accuracy of MLP for the

classification of severe wear tool conditions attain

100% accuracy. TCM prediction output on normal

tool use is always in the range -1 to -0.8 for 153 test

case on 9 parameters of machining. With these

results, the accuracy of TCM for the classification of

normal tool conditions attain 100% accuracy.

Unlike the results of validation on the use of

severe wear tools and normal tools, the validation of

TCM prediction for the use of moderate wear tools

have not reached 100% accuracy. In general, most

TCM conclusions are in the range -0.8 to 0.8. But in

cutting parameters with 0.08 mm/tooth feed, 750 rpm,

all test case samples were indicated severe wear

because they were in the range 0.8 to 1. In cutting

parameters 0.05 mm/tooth feed, 650 rpm, 9 of 17 test

case samples were indicated severe wear, even though

it was moderate wear.

Table 2: MSE of classifier during training phase and accuracy of classifier in testing phase and validation phase.

Training Phase Testing Phase Validation Phase

Train

Iteration

Epoch MSE Gradient Missed False

Alarm

Acuracy Missed False

Alarm

Acuracy

44 0,0402 0,0498 3,0% 2,5% 97,2% 0,6% 0,6% 99,4%

62 0,0376 0,00631 6,4% 2,3% 95,6% 2,8% 1,7% 97,8%

18 0,0485 0,0193 5,4% 4,5% 95,0% 0,0% 1,7% 99,2%

25 0,0411 0,036 8,5% 4,7% 93,3% 0,0% 0,0% 100,0%

50 0,0408 0,0112 4,7% 4,2% 95,6% 1,1% 0,6% 99,2%

53 0,0389 0,00834 3,2% 2,3% 97,2% 3,9% 0,0% 98,1%

43 0,0436 0,00771 8,9% 5,0% 93,3% 0,6% 0,0% 99,7%

18 0,0524 0,0138 7,2% 6,0% 93,3% 0,0% 0,0% 100,0%

41 0,0413 0,0093 5,4% 2,3% 96,1% 1,1% 1,7% 98,6%

45 0,0443 0,0221 6,5% 4,6% 94,4% 0,0% 0,0% 100,0%

Best 18 0,0376 0,00631 3,0% 2,3% 97,2% 0,0% 0,0% 100,0%

Average 40 0,04287 0,01838 5,9% 3,8% 95,1% 1,0% 0,6% 99,2%

Worse 62 0,0524 0,0498 8,9% 6,0% 93,3% 3,9% 1,7% 97,8%

Post-Processing for Three Class of Tool Wear Prognosis using Two Class ANN Classiﬁer based on Vibration of CNC Milling

273

Table 3: Detail of TCM prediction.

Spindle (rpm) 750 650 550 Tota

Feed (mm/tooth)

0,0

treatment 1 2 3 4 5 6 7 8 9

Predicted

severe 17 7 1 6 9 2 0 4 0

moderate 0 10 16 11 8 15 17 13 17

107

normal 0 0 0 0 0 0 0 0 0

Sub-total 17 17 17 17 17 17 17 17 17 153

Accuracy(%) 0,0

58,

94,

64,

47,

88,

76,

69,9

Table 4: Confussion matrix of TCM classifier.

True Class

severe moderate normal

Predicted Class

severe 153 46 0

moderate 0 107 0

normal 0 0 153

Accuracy 100% 69,9% 100% 89,98%

The detailed results of the TCM inference for the

9 cutting parameter treatments are shown in Table

III. The best accuracy of 100% on moderate wear

tool use is achieved for 0.02 mm/tooth feed. The

thicker the feed per tooth, the lower the inference

accuracy, and there tends to be misclassification as

the tool is severe wear. Likewise for spindle speed

cutting parameter. The faster the spindle turns, the

more accurate the inference is, and the prediction

error tends to misclassification as severe tool wear.

Detailed confusion matrix 3 × 3 TCM validation

using 3 levels of tool wear is shown in Table IV.

The TCM summary was obtained from TCM

predictions by Figure 6, Figure 7, and Figure 8. It

appears in TCM validation on use of severe wear

tool, that 153 out of 153 test cases indicated

precisely as a severe wear condition. Likewise in

TCM validation using normal tools, 153 out of 153

test cases are correctly indicated as normal tools. In

both classes there is no misclassification.

5 CONCLUSIONS

Validation of TCM predictions using moderate wear

tools, 107 out of 153 inferences are precisely

indicated as moderate wear tools, 46

missclassifications as broken tools, and no

missclassification as normal tools. Missclassification

that occurs in the validation of moderate wear test

cases is more conservative, because the actual tool

that are moderate wear are incorrectly indicated as

severe wear, and none are incorrectly indicated as

normal tool. This characteristic is safer (fail safe

design) for monitoring tool conditions, because no

case detection is missed. Overall TCM accuracy for

3 level of tool wear is 89.98% from 459 test cases.

Multi-layer perceptron that has been trained

using 2 classes; normal tools and severe wear tools,

with appropriate threshold and post-processing

settings can be used to classify 3 level of tool wear.

Using moving average smoothing with period 4,

threshold -0.8 for normal tool and 0.8 threshold for

severe wear tool, the multi-layer perceptron can

classify severe wear, moderate wear, and normal

tool condition with an accuracy of 89.98% of 459

test case of validation dataset.

REFERENCES

Ahmad, R., Kamaruddin, S., 2012. A review of condition-

based maintenance decision-making. Eur. J. Ind. Eng.

6, 519.

Arendra, A., Akhmad, S., Winarso, K., Herianto, H., 2020.

Investigating pump cavitation based on audio sound

signature recognition using artificial neural network. J.

Phys. Conf. Ser. 1569.

Arendra, A., Herianto, H., 2020. Pre-processing for

vibration signals features extraction and selection in

real time investigating of CNC tool wear. J. Phys.

Conf. Ser. 1569.

Barraza, S.M. del C., 2017. Unsupervised Feature

Learning Applied to Condition Monitoring. Lulea

University of Technology, Sewden.

Benkedjouh, T., Zerhouni, N., Rechak, S., 2017. Bearings

prognostics based on blind sources separation and

robust correlation analysis. In: ICINCO 2017 -

CESIT 2020 - International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies

274

Proceedings of the 14th International Conference on

Informatics in Control, Automation and Robotics.

SCITEPRESS – Science and Technology Publications,

Lda, pp. 658–663.

Casoli, P., 2019. A Methodology Based on

Cyclostationary Analysis for Fault Detection of

Hydraulic Axial Piston Pumps. energies 11.

Casoli, P., Pastori, M., Scolari, F., Rundo, M., 2019. A

Vibration Signal-Based Method for Fault

Identification and Classification in Hydraulic Axial

Piston Pumps. energies 12.

Cheng, Y., Zhu, H., Hu, K., Wu, J., Shao, X., Wang, Y.,

2019. Multisensory Data-Driven Health Degradation

Monitoring of Machining Tools by Generalized

Multiclass Support Vector Machine. IEEE Access 7,

47102–47113.

Goti, A., Oyarbide-zubillaga, A., Alberdi, E., Sanchez, A.,

2019. Optimal Maintenance Thresholds to Perform

Preventive Actions by Using Multi-Objective

Evolutionary Algorithms. Appl. Sci. 9, 1–13.

He, Y., Huang, M., Sun, W., 2019. Tool wear status

recognition based on Mahalanobis distance. J. Eng.

2019, 8802–8805.

Huang, Z., Zhu, J., Lei, J., Li, X., Tian, F., 2019. Tool

Wear Predicting Based on Multisensory Raw Signals

Fusion by Reshaped Time Series Convolutional

Neural Network in Manufacturing. IEEE Access 7,

178640–178651.

Jemielniak, K., Urbański, T., Kossakowska, J., Bombiński,

S., 2012. Tool condition monitoring based on

numerous signal features. Int J Adv Manuf Technol

59, 73–81.

Junior, P., D’Addona, D.M., Aguiar, P., Teti, R., 2018.

Dressing Tool Condition Monitoring through

Impedance-Based Sensors: Part 2—Neural Networks

and K-Nearest Neighbor Classifier Approach. sensors

18.

Lembke, B., 2019. Bearing Diagnosis using Fault Signal

Enhancing Techniques and Data-driven Classification.

Linköping University, Sweden.

Leturiondo, U., Salgado, O., Galar, D., 2016. Test rig

model development and validation for the diagnosis of

rolling element bearings. In: 14th IMEKO TC10

Workshop Technical Diagnostics New Perspectives in

Measurements, Tools and Techniques for System’s

Reliability, Maintainability and Safety. Milan, Italy,

pp. 46–49.

Liu, Y., Guo, C., Zhao, J., Xie, H., Sun, W., 2010.

Application of hierarchical model method on open

CNC system’s behavior reconstruction. In: ICINCO

2010 - Proceedings of the 7th International Conference

on Informatics in Control, Automation and Robotics.

SCITEPRESS – Science and Technology Publications,

Lda, pp. 172–175.

Mei, J., Luo, M., Guo, J., Li, H., Zhang, D., 2018.

Analytical Modeling, Design and Performance

Evaluation of Chatter-Free Milling Cutter with

Alternating Pitch Variations. IEEE Access 6, 32367–

32375.

Mishra, M., 2015. Model-based Prognostics for Prediction

of Remaining Useful Life. Luleå University of

Technology Sweden.

Murat, Z., Brezak, D., Augustin, G., Majetic, D., 2017.

Frequency domain analysis of acoustic emission

signals in medical drill wear monitoring. In:

BIOSIGNALS 2017 - 10th International Conference

on Bio-Inspired Systems and Signal Processing,

Proceedings; Part of 10th International Joint

Conference on Biomedical Engineering Systems and

Technologies, BIOSTEC 2017. SCITEPRESS –

Science and Technology Publications, Lda, pp. 173–

177.

Pappachan, B.K., Tjahjowidodo, T., Wijaya, T., 2017.

Event classification from sensor data using spectral

analysis in robotic finishing processes. In:

SENSORNETS 2017 - Proceedings of the 6th

International Conference on Sensor Networks.

SCITEPRESS – Science and Technology Publications,

Lda, pp. 80–86.

Prasetyo, T., Amar, S., Arendra, A., Zam Zami, M.K.,

2018. On-line Tool Wear Detection on DCMT070204

Carbide Tool Tip Based on Noise Cutting Audio

Signal using Artificial Neural Network. J. Phys. Conf.

Ser. 953.

Rmili, W., Ouahabi, A., Serra, R., Leroy, R., 2016. An

automatic system based on vibratory analysis for

cutting tool wear monitoring. Measurement 77, 117–

123.

Sakthivel, N.R., Nair, B.B., Elangovan, M., Sugumaran,

V., Saravanmurugan, S., 2014. Comparison of

dimensionality reduction techniques for the fault

diagnosis of mono block centrifugal pump using

vibration signals Engineering Science and Technology

, an International Journal Comparison of

dimensionality reduction techniques for the fault .

Eng. Sci. Technol. an Int. J. 17, 30–38.

Teti, R., Jemielniak, K., O’Donnell, G., Dornfeld, D.,

2010. Advanced monitoring of machining operations.

CIRP Ann. - Manuf. Technol. 59, 717–739.

Tobon-Mejia, D. a. a, Medjaher, K., Zerhouni, N., 2012.

CNC machine tool’s wear diagnostic and prognostic

by using dynamic Bayesian networks. Mech. Syst.

Signal Process. 28, 167–182.

Wei, P., Wang, H., 2019. Evaluation method of spindle

performance degradation based on VMD and random

forests. J. Eng. 2019, 8862–8866.

Wiharto, Kusnanto, H., Herianto, 2015. Performance

Analysis of Multiclass Support Vector Machine

Classification for Diagnosis of Coronary Heart

Diseases. Int. J. Comput. Sci. Appl. 5.

Wiharto, W., Kusnanto, H., Herianto, H., 2016.

Intelligence System for Diagnosis Level of Coronary

Heart Disease with K-Star Algorithm. Healthc.

Inform. Res. 22.

Yau, H.-T., Chen, J.-L., Yu, B.-R., Yang, T.-J., 2014.

Development of 3D Simulation System for Multi-Axis

Turn-Mill Machining. In: Proceedings of the 4th

International Conference on Simulation and Modeling

Methodologies, Technologies and Applications

Post-Processing for Three Class of Tool Wear Prognosis using Two Class ANN Classiﬁer based on Vibration of CNC Milling

275

(SIMULTECH-2014). SCITEPRESS – Science and

Technology Publications, Lda, pp. 717–724.

Zhou, C., Guo, K., Yang, B., Wang, H., Sun, J., Lu, L.,

2019. Singularity Analysis of Cutting Force and

Vibration for Tool Condition Monitoring in Milling.

IEEE Access 7, 134113–134124.

CESIT 2020 - International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies

276