Study of a Surface Roughness Modulation System for CNC End Milling
碩士 === 國立高雄第一科技大學 === 機械與自動化工程系碩士班 === 105 === The surface quality by the end milling process is affected by process parameters. In the past the process parameters are usually set according to suggestions of tooling companies, and thus it is difficult to control the surface quality or sometimes it t...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | zh-TW |
Published: |
2017
|
Online Access: | http://ndltd.ncl.edu.tw/handle/z6dbx8 |
id |
ndltd-TW-105NKIT0689017 |
---|---|
record_format |
oai_dc |
collection |
NDLTD |
language |
zh-TW |
format |
Others
|
sources |
NDLTD |
description |
碩士 === 國立高雄第一科技大學 === 機械與自動化工程系碩士班 === 105 === The surface quality by the end milling process is affected by process parameters. In the past the process parameters are usually set according to suggestions of tooling companies, and thus it is difficult to control the surface quality or sometimes it takes a lot of time to achieve good surface quality by trial and error. Therefore, this research aims at developing the mechanism of predicting the surface quality for end milling processes and also the modification method of the process parameters to achieve the desired surface quality.
The surface roughness modulation system mainly consists of two subsystems, i.e., the surface roughness prediction model and the surface roughness modulation mechanism. The surface roughness prediction model was established using a backpropagation neural network (NN), in which the inputs are spindle speed, feedrate, depth of cut, cutting force features and a bias -1. The input -1 provides an additional degree of freedom for the threshold, so that the training process of the neural network is accelerated, while the cutting force features consider the RMS values, vibration amplitudes and mean values of five different cutting forces, which are the cutting forces in the X and Y directions, the resultant ones in the XY planes, the ones in the Z direction, the ones in the X, Y, and Z directions, and the resultant ones in the XY planes combined with the ones in the Z direction. As for the neural network structure, one and two hidden layers are considered, and the number of the neurons in each hidden layer are from 1 to 10. The output of the NN model is the surface roughness. The above variation in the NN structure results in 300 different combinations. Then the MATLAB NN toolbox is utilized to conduct NN training and the best NN structure is the one with the MEAN cutting force feature for the resultant forces in the XY planes combined with the ones in the Z direction, the neuron number 6-10-1 in each layer. This best NN model, with the training accuracy 95.21% and the testing accuracy is 92.72% is chosen as the surface roughness prediction model.
As for developing the surface roughness modulation mechanism, a database is first established and then a surface roughness modulation system is built up based on the database. To establish the database, the first step is to set the configuration of machining parameters, i.e., fix spindle speed and cutting depth but only adjust the feedrate. The second step is to establish a neural network model which utilizes spindle speed, feedrate, and depth of cut (the features of cutting forces is not included here) to predict surface roughness. The resultant NN model with a structure of 4-7-7-1 possesses a training accuracy of 93.52% and a testing accuracy of 92.36%. The third step is to process the machining-parameter configuration and then use the NN model to predict the corresponding surface roughness. In the configuration, a set of a data group is arranged to possess the same spindle speed and depth of cut, but the modulated feedrate and the corresponding surface roughness change. Next the surface roughness modulation system adopts two strategies, i.e., automatic and manual modulation. The automatic modulation uses nonlinear regression analysis and neural networks, respectively, to establish the relationship between the inputs (i.e., spindle speed, feedrate, depth of cut, and desired surface roughness change) and the output (i.e., feedrate adjustment). The average accuracy of the nonlinear regression model is 90.02%, while that of the neural network model model with the structure 5-3-1 is 88.97%. Here the data for training the neural networks are divided into the training and validation sets, but those for the regression analysis are not divided. Therefore, the neural network model is adopted for the automatic surface roughness modulation system. As for the manual modulation, it uses machining parameters to predict the surface roughness by a neural network model. If the surface roughness is worse than the preset value, the feedrate is adjusted to be 10mm/min lower until achiving the preset roughness. Simulation shows that both of the automatic and manual modulation methods could achieve the desired surface roughness. The experimental verification is left as one of the future works.
|
author2 |
LI, CHEN-JUNG |
author_facet |
LI, CHEN-JUNG TSAI, PO-JUNG 蔡博戎 |
author |
TSAI, PO-JUNG 蔡博戎 |
spellingShingle |
TSAI, PO-JUNG 蔡博戎 Study of a Surface Roughness Modulation System for CNC End Milling |
author_sort |
TSAI, PO-JUNG |
title |
Study of a Surface Roughness Modulation System for CNC End Milling |
title_short |
Study of a Surface Roughness Modulation System for CNC End Milling |
title_full |
Study of a Surface Roughness Modulation System for CNC End Milling |
title_fullStr |
Study of a Surface Roughness Modulation System for CNC End Milling |
title_full_unstemmed |
Study of a Surface Roughness Modulation System for CNC End Milling |
title_sort |
study of a surface roughness modulation system for cnc end milling |
publishDate |
2017 |
url |
http://ndltd.ncl.edu.tw/handle/z6dbx8 |
work_keys_str_mv |
AT tsaipojung studyofasurfaceroughnessmodulationsystemforcncendmilling AT càibóróng studyofasurfaceroughnessmodulationsystemforcncendmilling AT tsaipojung shùkòngduānxiǎnxuēbiǎomiàncūcāodùdiàobiànxìtǒngzhīyánjiū AT càibóróng shùkòngduānxiǎnxuēbiǎomiàncūcāodùdiàobiànxìtǒngzhīyánjiū |
_version_ |
1719150310456295424 |
spelling |
ndltd-TW-105NKIT06890172019-05-15T23:32:32Z http://ndltd.ncl.edu.tw/handle/z6dbx8 Study of a Surface Roughness Modulation System for CNC End Milling 數控端銑削表面粗糙度調變系統之研究 TSAI, PO-JUNG 蔡博戎 碩士 國立高雄第一科技大學 機械與自動化工程系碩士班 105 The surface quality by the end milling process is affected by process parameters. In the past the process parameters are usually set according to suggestions of tooling companies, and thus it is difficult to control the surface quality or sometimes it takes a lot of time to achieve good surface quality by trial and error. Therefore, this research aims at developing the mechanism of predicting the surface quality for end milling processes and also the modification method of the process parameters to achieve the desired surface quality. The surface roughness modulation system mainly consists of two subsystems, i.e., the surface roughness prediction model and the surface roughness modulation mechanism. The surface roughness prediction model was established using a backpropagation neural network (NN), in which the inputs are spindle speed, feedrate, depth of cut, cutting force features and a bias -1. The input -1 provides an additional degree of freedom for the threshold, so that the training process of the neural network is accelerated, while the cutting force features consider the RMS values, vibration amplitudes and mean values of five different cutting forces, which are the cutting forces in the X and Y directions, the resultant ones in the XY planes, the ones in the Z direction, the ones in the X, Y, and Z directions, and the resultant ones in the XY planes combined with the ones in the Z direction. As for the neural network structure, one and two hidden layers are considered, and the number of the neurons in each hidden layer are from 1 to 10. The output of the NN model is the surface roughness. The above variation in the NN structure results in 300 different combinations. Then the MATLAB NN toolbox is utilized to conduct NN training and the best NN structure is the one with the MEAN cutting force feature for the resultant forces in the XY planes combined with the ones in the Z direction, the neuron number 6-10-1 in each layer. This best NN model, with the training accuracy 95.21% and the testing accuracy is 92.72% is chosen as the surface roughness prediction model. As for developing the surface roughness modulation mechanism, a database is first established and then a surface roughness modulation system is built up based on the database. To establish the database, the first step is to set the configuration of machining parameters, i.e., fix spindle speed and cutting depth but only adjust the feedrate. The second step is to establish a neural network model which utilizes spindle speed, feedrate, and depth of cut (the features of cutting forces is not included here) to predict surface roughness. The resultant NN model with a structure of 4-7-7-1 possesses a training accuracy of 93.52% and a testing accuracy of 92.36%. The third step is to process the machining-parameter configuration and then use the NN model to predict the corresponding surface roughness. In the configuration, a set of a data group is arranged to possess the same spindle speed and depth of cut, but the modulated feedrate and the corresponding surface roughness change. Next the surface roughness modulation system adopts two strategies, i.e., automatic and manual modulation. The automatic modulation uses nonlinear regression analysis and neural networks, respectively, to establish the relationship between the inputs (i.e., spindle speed, feedrate, depth of cut, and desired surface roughness change) and the output (i.e., feedrate adjustment). The average accuracy of the nonlinear regression model is 90.02%, while that of the neural network model model with the structure 5-3-1 is 88.97%. Here the data for training the neural networks are divided into the training and validation sets, but those for the regression analysis are not divided. Therefore, the neural network model is adopted for the automatic surface roughness modulation system. As for the manual modulation, it uses machining parameters to predict the surface roughness by a neural network model. If the surface roughness is worse than the preset value, the feedrate is adjusted to be 10mm/min lower until achiving the preset roughness. Simulation shows that both of the automatic and manual modulation methods could achieve the desired surface roughness. The experimental verification is left as one of the future works. LI, CHEN-JUNG 李振榮 2017 學位論文 ; thesis 113 zh-TW |