1. bookAHEAD OF PRINT
Detalles de la revista
License
Formato
Revista
eISSN
2444-8656
Primera edición
01 Jan 2016
Calendario de la edición
2 veces al año
Idiomas
Inglés
Acceso abierto

Prediction of surface quality in end milling based on modified convolutional recurrent neural network

Publicado en línea: 16 Aug 2022
Volumen & Edición: AHEAD OF PRINT
Páginas: -
Recibido: 18 Jan 2022
Aceptado: 15 May 2022
Detalles de la revista
License
Formato
Revista
eISSN
2444-8656
Primera edición
01 Jan 2016
Calendario de la edición
2 veces al año
Idiomas
Inglés
Introduction

‘Industry 4.0’ is significantly considered as the fourth industrial revolution [1], in which smart factories and intelligent manufacturing play a significant part. Besides new concepts of Internet of Things (IoT) [2], digital twins (DT) [3] and cloud manufacturing (CM) [4], the fundamental aspects of industry still heavily rely on traditional production procedures [5] such as casting, moulding, welding and machining. Since machining is vital for precision engineering and complex workpieces, much effort has been directed at improving the performance of NC machining. The quality of the milled surface affects the performance of the affiliated workpiece, since it greatly determines the precision of the geometry and duration of service time. To develop advanced manufacturing towards intelligent factories, surface quality prediction is of great research focus [6]. Comprehensive related approaches and studies have been reported on surface profile and the roughness of machined products [7, 8]. Generally, many features, such as cutting condition, tool status, material properties and process parameters (tool geometry, cut depth, feed rate and spindle) are considered to have a significant influence on the surface roughness. As the determination mechanism of surface roughness is complicated, no simple closed-form model could accurately predict the surface finish.

Ending milling, which accounts for a large proportion of milling, is universally used in moulding, aerospace and automobile industries [9]. The surface quality is not only an important index to measure whether the work-piece meets the requirements but also the primary goal of monitoring the machining process and parameters’ optimisation. Therefore, it is essential to predict the quality before actual machining is performed, to decrease the time for boring parameter trials and corrections, and reduce the cost of the manpower. Lee et al. [10] proposed a high-speed milling surface roughness simulation approach, which adopted acceleration signals instead of cutting force signals for studying high-speed milling surface roughness. In the research, a statistical milling model was established for analysing the influence of spindle deformation and vibration on roughness. Mizugaki [11] and Kim [12], respectively, discussed the effect of the blunt radius of the blade on the surface roughness in the study of the tool path of the milling process. Then, Singh et al. [13] studied the bearing steel with higher hardness and used tools with different radii and bevel angles. The experimental results show that the feed rate has the most significant influence on the surface quality, while the bevel angle has the least influence on the machined surface. Quinsat et al. [14] believed that the processing strategy and parameters are related to the surface quality. The process parameters are obtained from the angle and the required surface quality, and the machining processing is simulated and tested. Results show that the importance of parameter design did achieve the required processing surface through parameters set strategy. Different cutting conditions make different influencing facts, and thus no closed-form model could achieve accurate predicting, because different materials and different parameters, as well as different machine tools, can alter the cutting condition. Thus, direct statistical simulation or single effect analysis might not fulfil the requirement for predicting the surface quality.

Other researchers used evolutions, artificial intelligent algorithms or deep network for surface quality prediction. Zain et al. [15] used genetic algorithm to analyse the radial clearance angle of the milling cutter and considered the influence of feed and cutting depth on roughness, and finally found the optimal surface under the combination of high speed, low feed rate and radial clearance angle. Cosres et al. [16] used a non-contact displacement measurement system to record the movement of the tool during processing and the problem of tool vibration at high frequencies, and derived the surface topography after processing. Using traditional inspection equipment to inspect the surface quality of processed workpieces takes a lot of time, and it is impossible to predict the surface quality during processing. To be able to predict the surface quality during the machining process and adjust the machining process parameters in time to improve the machining quality, an intelligent detection method is urgently needed. Sahith Reddy Madara et al. [17] used an artificial neural network to model and predict surface roughness in abrasive waterjet cutting of Kevlar 49. Girish Kant et al. [18] developed an optimisation model by coupling the ANN and GA. In addition, evolution algorithm and artificial intelligence methods make great application prospects.

At present, in the machining process, various digital signals are generally obtained directly through sensors, but the collected data are always contaminated with noises, which seriously hinders the identification of the actual wear status of the tool, and thus could not reflect the surface quality. In this work, a modified convolutional recurrent neural network (CRNN) is applied to the prediction of surface quality. First, the validated features of milling force data in the machining process are extracted based on the proposed artificial network model. Then, a modified CRNN model is constructed by combining residual neural network and bidirectional long- and short-term memory as well as attention mechanism. The weight coefficients in the model are optimised according to the change of loss function and directional propagation principle, which greatly improves the effectiveness of the proposed model. Finally, the real experiment is carried out on a 5-axis milling centre to validate the proposed approach.

Construction of modified CRNN model
The classic CRNN network

Convolutional neural network (CNN) and recurrent neural network (RNN) are the main components of CRNN. CNN consists of multiple convolutional layers, pooling layers and fully connected layers. To deal with the enormous calculation, gradient loss and over-fitting involved during the back propagation process, a rectified linear unit (ReLU) function will be used. This could provide rapid convergence speed. Then RNNs are adopted for modelling dynamic changes in time series. The details about classic CRNN can be found in Li et al. [19]; this study used CRNN for predicting combustion states of a rotary kiln. The classical CRNN architecture combines the advantages of deep CNN and RNN, and gives a good result; further, it shows that CRNN makes for a great application prospect. Also, CRNN network also has the characteristics of both CNN and RNN networks.

The modified CRNN network

Due to the over-fitting phenomenon in the deep learning network in CRNN in the feature extraction process, and the insufficient recognition of the full-position context information by the convolutional network, based on the original CRNN, residual neural network, the bidirectional long- and short-term memory and attention mechanism are applied to build a modified CRNN model.

Components of modified CRNN

Most conventional CNN models pile up the convolutional layers to deepen the network to elevate the accuracy of the recognition. However, as the network structure deepens (resulting in redundancy of the convolutional layers), problems come to light. On the one hand, gradient loss or disappearance results in the inefficiency of updating the gradient to the previous network layers through back error propagation, causing parameters to not be updated. On the other hand, the model accuracy appears saturated and the accuracy declines rapidly, which means that the network degrades. The deep residual network is proposed to solve these problems. In this work, DenseNet is mainly introduced, discussed and applied to the CRNN framework. The CNN network performs well on discovering the local spatial features, while it cannot reflect the temporal relation information of the signals in time series.

The CRNN structure proposed in this paper aims to increase the depth of CNN network structure based on the framework of machining surface quality recognition algorithm, so as to improve its capability of deep feature extraction. Meanwhile, the structure of residual network is added to avoid the phenomenon of over-fitting in the process of feature extraction of the deep network. At the same time, BILSTM and attention mechanism are combined to extract deep features, which makes up for the lack of a general convolutional network to recognise the complete location context information. Eventually, we integrate features through the fully connected layer and output the final status recognition result after softmax. In this paper, the machined surface quality is divided into three categories according to the value range of surface roughness value: good, medium and poor. The whole CRNN model framework is shown in Table 1.

The structure of modified CRNN

Layers Size DenseNet(4_2)

Convolution 5000 Conv 1*7, stride = 2
Pooling 2500 Maxpool 1*3, stride = 2
Dense Block 1 3000 (1*2 conv, 1*1 conv)*4
Transition 3000 1*1 conv
Block 1500 1*2 average pool, stride = 2
BILSTM 725 BiLSTM, unit = 128
Attention 725 Atten = 128
Global pooling 128 Average pool
Fully connect 64 Fully connected
Output 3 Softmax

The final network structure is shown in Figure 1. The inputs are three-direction cutting force signals. Since the sampling frequency in this experiment is 2500 Hz, the data collected in the experiment are enhanced through down-sampling to obtain more data samples. Finally, the input dimension of the network is set as (10,000,1).

Fig. 1

Diagram of machining surface quality prediction model based on modified CRNN network

The input passes through the convolutional layer, pooling layer, DenseNet residual layer and BILSTM attention layer to complete the identification of machining surface quality. The structure of the whole network is shown in Table 1. The convolution kernel size is reduced, which can, on the one hand, improve the robustness of the network, and on the other hand, reduce the computational cost. Meanwhile, to adapt to the number of the output channels in the convolutional layer, a constant factor ranging (0,1) is added after the output of the dense block to adjust the number of output channels.

Training of modified CRNN

After the network structure is determined, the process of surface quality recognition based on CRNN has also been determined. First, the signal of the original milling force is down-sampled to obtain the expanded data. Then, the data are normalised. Finally, the data are divided into training set and testing set with the ratio of 8:2, the training samples are input into the original CRNN network with four training control mechanisms and the training network enables it to have the fitting ability of machining surface quality recognition. An early stopping mechanism ensures the network will not overfit. Eventually, recognition ability and effect of the trained CRNN network are verified on testing set. The process flow of the model is shown in Figure 2.

Fig. 2

Flow diagram of machining surface quality prediction model based on modified CRNN network

After the prediction of the machining surface quality CRNN model and the training process, the weights in the model will be updated and optimised through the loss function and the principle of direction propagation, resulting in the gradual improvement of recognition ability. Figure 3 shows the changes of loss value and accuracy of the entire network in the training and testing sets. With the training process, loss value decreases continuously on both the training and testing sets, while the accuracy rate increases. Finally, the convergence is realised, and the prediction accuracy of the CRNN model on the verification set reaches 98.35%, which confirms the applicability and accuracy of the CRNN network proposed in this paper.

Fig. 3

CRNN model indicators for surface quality prediction

To verify the ability of the machining surface quality recognition and the effect of different CRNN deep networks, the DenseNet and BILSTM structure in CRNN deep network are separated and trained, respectively. The input is uniformly set as the force signal in the machining process, the output is set as three categories of surface quality and the test set is used for verification. The result of loss function on the testing set is shown in Figure 4, while the result of accuracy on testing set is shown in Figure 5. As is shown in Figures 4 and 5, in contrast with the rapid convergence and high accuracy of the prediction results of the CRNN, the loss of the DenseNet model on the testing set is more volatile, and the accuracy on the testing set is lower than that on the CRNN, although the DenseNet model can eventually converge. BILSTM is obviously not as effective at this task as the other two network structures.

Fig. 4

The loss function of different algorithms varies with the number of iterations

Fig. 5

The accuracy curve of different algorithms on the verification set

According to the variation trend of loss value in Figure 5, it can be seen that DenseNet network can learn and extract the features of signals in the spatial domain, while BILSTM can extract the features of input signals in the temporal dimension and fit the correlation between various features through the attention mechanism. However, only these two parts of the network structure will cause the problem of insufficient model fitting ability, and thus it is necessary to combine the two effectively. The training results of the three networks are compared as shown in Table 2.

Comparison of prediction accuracy and operation time of different network

Structure Accuracy Iterative convergence
Proposed CRNN DenseNet(4_2) BILSTM 98.35% 97.68% 72.96% 15 18 22

The proposed CRNN can learn more related characteristics compared with ordinary multi-layer neural networks. At the same time, full correlation analysis is carried out on the extracted features through the BILSTM network with attention mechanism, so as to realise the accurate recognition of the machining surface quality. Based on the data in Table 2, the effectiveness of the proposed CRNN network on the surface quality prediction can be verified.

Experimental results and discussion
Equipment setup

This work aimed to investigate the effect of cutting parameters and conditions on surface roughness parameters, such as cutting tool, stiffness of machine tool and material. In our experiment, the workpiece material was 42CrMo steel with a rectangular shape of 65 mm × 140 mm × 40 mm. The steel has good corrosion resistance and high hardness of 30 HRC. Therefore, its surface quality is challenging to guarantee during machining process. Table 3 shows the chemical composition of 42CrMo steel and Figure 6 shows the shape of the workpiece.

Fig. 6

42CrMo steel workpiece

Workpiece material composition

Composition C Si Mn P S Cr Mo
Wt(%) 0.38–0.43 0.15–0.35 0.75–1.00 ≤ 0.035 ≤ 0.040 0.80–1.10 0.15–0.25

As shown in Figure 7, a series of dry milling experiments were conducted on Starrag LX051, which has a maximum of 18,000 rpm, rated torque of 181 Nm, positioning accuracy of 0.004 mm and repeat positioning accuracy of 0.002 mm. The cutting tool used for the experiments is a detachable end mill. The diameter of the shank is 16 mm. The insert on the shank is square-shouldered milling cutters (PKT11T7308-PM). During milling, a Spike 1.2 rotary force measuring instrument was used to measure the three-way force. The roughness was measured by the SJ-210 surface roughness tester, whose accuracy can reach 0.001 μm.

Fig. 7

Experimental setup

The orthogonal experiment method [20] is used to replace the enormous amount times of tests with only a partial part while still expressing the whole situation. This investigation is carried out upon surface roughness with different machining parameter combinations. To improve the experimental efficiency, this research work chooses the spindle speed, the axial cutting depth and the feed per tooth as the influencing factors, and each factor has three levels. Three levels were specified for each factor, as indicated in Table 4. The orthogonal array chosen was L7, which has seven rows corresponding to the number of parameter combinations, with three columns at three levels as shown in Table 5, in which A, B and C represent the machining parameters. The surface morphology of five groups is described in Table 6, and pictographic representation of the surface morphology is also provided in Figure 8.

Fig. 8

Surface topography samples

Experiment factor and level

Level Factor

Spindle speed S (rpm) Cutting depth ap (mm) Feed per tooth fz (mm/tooth)

1 3200 0.5 0.0625
2 6400 1.5 0.125
3 8000 2 0.25

Designed orthogonal experiment

Index A B C

1 1 1 1
2 1 2 2
3 1 3 3
4 2 1 2
5 2 2 3
6 2 3 1
7 3 1 3

Process parameters of cutting experiment samples

Index Feed per tooth (mm) Surface topography Surface roughness

7 0.05 Figure 8(A) 0.567
2 0.0625 Figure 8(B) 0.612
3 0.125 Figure 8(C) 0.706
8 0.2 Figure 8(D) 1.025
5 0.25 Figure 8(E) 1.395
Analysis of surface roughness results

The range of processed surface quality can be accurately predicted by the constructed deep network, and in real application scenarios, it is often necessary to accurately identify and predict the surface roughness more accurately, so as to accurately obtain the quality results of each process to ensure that each process is within the accuracy requirements. Based on the quality recognition model in the previous section, using the extracted internal features to build a regression model, mapping output to surface roughness value, that is, the input is the milling force signal during machining, and the output is the surface roughness prediction result. The specific process is shown in Figure 9. It involves replacing the softmax structure of the last layer of the CRNN network with a regression model to build a surface roughness prediction model during processing and using MSE mean square error loss function as objective function.

Fig. 9

Surface roughness prediction model

When conducting the surface roughness prediction experiment, since it is necessary to predict the quality of the workpiece during the entire processing process, the training and test sets need to use a complete processing experiment data individually, and the roughness prediction model in this section is based on the previous one. For the classification model in this section, the model parameters before the softmax layer of the model in the previous section are migrated as the trained weights. Only three sets of data are used to fine-tune the regression model. The experiment selects three sets of data of 1, 2 and 7 as the training set, and four sets of data of 3, 4, 5 and 6 as the test set to verify the prediction effect of the model. The abscissa represents the number of processing cycles, and the ordinate represents the surface quality value. The real surface quality value after secondary processing is used as the standard, and the result of comparison with the model output is shown in Figure 10.

Fig. 10

Comparison of CRNN deep network model prediction value and experimental value

The results show that even if there is a certain error between the model prediction and the real situation, the CRNN model proposed in this paper can meet the repeatability and accuracy requirements of the surface quality prediction in the general processing process.

Conclusion

A modified CRNN was proposed to predict the surface quality of end milling. The proposed CRNN network could solve the over-fitting problem during the feature extraction process through optimisation of the convergence speed and loss function. The predicted values were compared with the actual experimental results, and the predicted values demonstrate a good accordance with the real cutting. Also, an accuracy rate of 98.35% could be achieved, which implies that, given our end milling circumstance, the experimental procedures adopted in the present study allowed us to predict the surface quality with higher precision and efficiency.

Fig. 1

Diagram of machining surface quality prediction model based on modified CRNN network
Diagram of machining surface quality prediction model based on modified CRNN network

Fig. 2

Flow diagram of machining surface quality prediction model based on modified CRNN network
Flow diagram of machining surface quality prediction model based on modified CRNN network

Fig. 3

CRNN model indicators for surface quality prediction
CRNN model indicators for surface quality prediction

Fig. 4

The loss function of different algorithms varies with the number of iterations
The loss function of different algorithms varies with the number of iterations

Fig. 5

The accuracy curve of different algorithms on the verification set
The accuracy curve of different algorithms on the verification set

Fig. 6

42CrMo steel workpiece
42CrMo steel workpiece

Fig. 7

Experimental setup
Experimental setup

Fig. 8

Surface topography samples
Surface topography samples

Fig. 9

Surface roughness prediction model
Surface roughness prediction model

Fig. 10

Comparison of CRNN deep network model prediction value and experimental value
Comparison of CRNN deep network model prediction value and experimental value

Experiment factor and level

Level Factor

Spindle speed S (rpm) Cutting depth ap (mm) Feed per tooth fz (mm/tooth)

1 3200 0.5 0.0625
2 6400 1.5 0.125
3 8000 2 0.25

Comparison of prediction accuracy and operation time of different network

Structure Accuracy Iterative convergence
Proposed CRNN DenseNet(4_2) BILSTM 98.35% 97.68% 72.96% 15 18 22

Designed orthogonal experiment

Index A B C

1 1 1 1
2 1 2 2
3 1 3 3
4 2 1 2
5 2 2 3
6 2 3 1
7 3 1 3

Process parameters of cutting experiment samples

Index Feed per tooth (mm) Surface topography Surface roughness

7 0.05 Figure 8(A) 0.567
2 0.0625 Figure 8(B) 0.612
3 0.125 Figure 8(C) 0.706
8 0.2 Figure 8(D) 1.025
5 0.25 Figure 8(E) 1.395

Workpiece material composition

Composition C Si Mn P S Cr Mo
Wt(%) 0.38–0.43 0.15–0.35 0.75–1.00 ≤ 0.035 ≤ 0.040 0.80–1.10 0.15–0.25

The structure of modified CRNN

Layers Size DenseNet(4_2)

Convolution 5000 Conv 1*7, stride = 2
Pooling 2500 Maxpool 1*3, stride = 2
Dense Block 1 3000 (1*2 conv, 1*1 conv)*4
Transition 3000 1*1 conv
Block 1500 1*2 average pool, stride = 2
BILSTM 725 BiLSTM, unit = 128
Attention 725 Atten = 128
Global pooling 128 Average pool
Fully connect 64 Fully connected
Output 3 Softmax

Liu Yue. The State, Industrial Organization and Industrial Development: Establishment and development of China Machine Tool Industry Association (1988–2016). Master's Thesis, Central China Normal University, 2020. LiuYue The State, Industrial Organization and Industrial Development: Establishment and development of China Machine Tool Industry Association (1988–2016) Master's Thesis, Central China Normal University 2020 Search in Google Scholar

Behin Elahi, Sadegh Amiri Tokaldany. Application of Internet of Things-aided simulation and digital twin technology in smart manufacturing. Advances in Mathematics for Industry 4.0, 2021:335–359. BehinElahi Sadegh AmiriTokaldany Application of Internet of Things-aided simulation and digital twin technology in smart manufacturing Advances in Mathematics for Industry 4.0 2021 335 359 10.1016/B978-0-12-818906-1.00015-2 Search in Google Scholar

Karl Hribernik, Giacomo Cabri, Federica Mandreoli, Gregoris Mentzas. Autonomous context-aware adaptive Digital Twins-State of the art and roadmap. Computers in Industry, 133(2021), 103508. HribernikKarl CabriGiacomo MandreoliFederica MentzasGregoris Autonomous context-aware adaptive Digital Twins-State of the art and roadmap Computers in Industry 133 2021 103508 10.1016/j.compind.2021.103508 Search in Google Scholar

Chao Liu, Ziwei Su, Xun Xu, Yuqian Lu. Service-oriented industrial internet of things gateway for cloud manufacturing. Robotics and Computer-Integrated Manufacturing, 73(2022), 102217. LiuChao SuZiwei XuXun LuYuqian Service-oriented industrial internet of things gateway for cloud manufacturing Robotics and Computer-Integrated Manufacturing 73 2022 102217 10.1016/j.rcim.2021.102217 Search in Google Scholar

Chanbeom Bak, Abhishek Ghosh Roy, Hungsun Son. Quality prediction for aluminium diecasting process based on shallow neural network and data feature selection technique. CIRP Journal of Manufacturing Science and Technology, 33(2021):327–338. BakChanbeom RoyAbhishek Ghosh SonHungsun Quality prediction for aluminium diecasting process based on shallow neural network and data feature selection technique CIRP Journal of Manufacturing Science and Technology 33 2021 327 338 10.1016/j.cirpj.2021.04.001 Search in Google Scholar

Chen Lu. Study on prediction of surface quality in machining process. Journal of Materials Processing Technology, 205(2008):439–450. LuChen Study on prediction of surface quality in machining process Journal of Materials Processing Technology 205 2008 439 450 10.1016/j.jmatprotec.2007.11.270 Search in Google Scholar

Wang Hongxiang, Dong Shen, Li Dan, Liang Feng. Effects of Optimum Controlled Vibration on Ultraprecision Machined Surface Quality. Chinese journal of mechanics, 11(4):452–455. (in Chinese) WangHongxiang DongShen LiDan LiangFeng Effects of Optimum Controlled Vibration on Ultraprecision Machined Surface Quality Chinese journal of mechanics 11 4 452 455 (in Chinese) Search in Google Scholar

Hong Quan, Wang Guicheng. The status and development of machined surface integrity in precision machining. Modern Manufacturing Engineering, 2004(8):12–15. HongQuan WangGuicheng The status and development of machined surface integrity in precision machining Modern Manufacturing Engineering 2004 8 12 15 Search in Google Scholar

Peter Michalik, Jozef Zajac, Michal Hatala, Dusan Mital, Veronika Fecova. Monitoring surface roughness of thin-walled components from steel C45 machining down and up milling. Measurement, 58(2014):416–428. MichalikPeter ZajacJozef HatalaMichal MitalDusan FecovaVeronika Monitoring surface roughness of thin-walled components from steel C45 machining down and up milling Measurement 58 2014 416 428 10.1016/j.measurement.2014.09.008 Search in Google Scholar

Ki Yong Lee, Myeong Chang Kang, Yung Ho Jeong, Deuk Woo Lee, Jeong Suk Kim. Simulation of surface roughness and profile in high-speed end milling. Journal of Materials Processing Tech., 2001, 113(1):410–415. LeeKi Yong KangMyeong Chang JeongYung Ho LeeDeuk Woo KimJeong Suk Simulation of surface roughness and profile in high-speed end milling Journal of Materials Processing Tech 2001 113 1 410 415 10.1016/S0924-0136(01)00697-5 Search in Google Scholar

Y. Mizugaki, M. Hao, K. Kikkawa, T. Nakagawa. Geometric Generating Mechanism of Machined Surface by Ball-nosed End Milling. CIRP Annals – Manufacturing Technology, 2001, 50(1):69–72. MizugakiY. HaoM. KikkawaK. NakagawaT. Geometric Generating Mechanism of Machined Surface by Ball-nosed End Milling CIRP Annals – Manufacturing Technology 2001 50 1 69 72 10.1016/S0007-8506(07)62073-3 Search in Google Scholar

B.H. Kim, C.N. Chu. Texture prediction of milled surfaces using texture superposition method. Computer-Aided Design, 1999, 31(8):485–494. KimB.H. ChuC.N. Texture prediction of milled surfaces using texture superposition method Computer-Aided Design 1999 31 8 485 494 10.1016/S0010-4485(99)00045-7 Search in Google Scholar

Dilbag Singh, P. Venkateswara Rao. A surface roughness prediction model for hard turning process. The International Journal of Advanced Manufacturing Technology, 2007, 32(11–12):1115–1124. SinghDilbag Venkateswara RaoP. A surface roughness prediction model for hard turning process The International Journal of Advanced Manufacturing Technology 2007 32 11–12 1115 1124 10.1007/s00170-006-0429-2 Search in Google Scholar

Y. Quinsat, L. Sabourin, C. Lartigue. Surface topography in ball end milling process: Description of a 3D surface roughness parameter. Journal of Materials Processing Tech., 2008, 195(1–3):135–143. QuinsatY. SabourinL. LartigueC. Surface topography in ball end milling process: Description of a 3D surface roughness parameter Journal of Materials Processing Tech 2008 195 1–3 135 143 10.1016/j.jmatprotec.2007.04.129 Search in Google Scholar

Azlan Mohd Zain, Habibollah Haron, Safian Sharif. Application of GA to optimize cutting conditions for minimizing surface roughness in end milling machining process. Expert Systems With Applications, 2010, 37(6):4650–4659. ZainAzlan Mohd HaronHabibollah SharifSafian Application of GA to optimize cutting conditions for minimizing surface roughness in end milling machining process Expert Systems With Applications 2010 37 6 4650 4659 10.1016/j.eswa.2009.12.043 Search in Google Scholar

Jean Philippe Costes, Vincent Moreau. Surface roughness prediction in milling based on tool displacements. Journal of Manufacturing Processes, 2011, 13(2):133–140. CostesJean Philippe MoreauVincent Surface roughness prediction in milling based on tool displacements Journal of Manufacturing Processes 2011 13 2 133 140 10.1016/j.jmapro.2011.02.003 Search in Google Scholar

Sahith Reddy Madara, Swaroop Ramaswamy Pillai, Chithirai Pon Selvan M, Jens Van heirle. Modelling of surface roughness in abrasive waterjet cutting of Kevlar 49 composite using artificial neural network. Materials Today: Proceedings, 2020(46), 1:1–8. MadaraSahith Reddy PillaiSwaroop Ramaswamy Chithirai Pon SelvanM Van heirleJens Modelling of surface roughness in abrasive waterjet cutting of Kevlar 49 composite using artificial neural network Materials Today: Proceedings 2020 46 1 1 8 10.1016/j.matpr.2020.02.868 Search in Google Scholar

Girish Kant, Kuldip Singh Sangwan. Predictive Modelling and Optimization of Machining Parameters to Minimize Surface Roughness using Artificial Neural Network Coupled with Genetic Algorithm. Procedia CIRP, 2015, 31:453–458. KantGirish SangwanKuldip Singh Predictive Modelling and Optimization of Machining Parameters to Minimize Surface Roughness using Artificial Neural Network Coupled with Genetic Algorithm Procedia CIRP 2015 31 453 458 10.1016/j.procir.2015.03.043 Search in Google Scholar

Tao Li, Zhenting Zhang, Hua Chen. Predicting the combustion state of rotary kilns using a Convolutional Recurrent Neural Network. Journal of Process Control, 2019, 84:207–214. LiTao ZhangZhenting ChenHua Predicting the combustion state of rotary kilns using a Convolutional Recurrent Neural Network Journal of Process Control 2019 84 207 214 10.1016/j.jprocont.2019.10.009 Search in Google Scholar

Xin Zhang, Jian wei Ma, Zhen yuan Jia, De ning Song. Machining parameter optimisation for aviation aluminium-alloy thin-walled parts in high-speed milling. International Journal of Machining and Machinability of Materials, 2018. XinZhang Jianwei Ma Zhenyuan Jia Dening Song Machining parameter optimisation for aviation aluminium-alloy thin-walled parts in high-speed milling International Journal of Machining and Machinability of Materials 2018 10.1504/IJMMM.2018.090546 Search in Google Scholar

Artículos recomendados de Trend MD

Planifique su conferencia remota con Sciendo