應用卷積神經網路於CNC攻牙之刀具狀態監測

Application of Convolutional Neural Network on Monitoring Tool Conditions in CNC Tapping

呂定憲(Ding-Shian Liu)

指導教授 ： 楊宏智

國立臺灣大學/工學院/機械工程學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201902830

摘要

在工業4.0與智慧製造的浪潮下，工具機逐漸由自動化邁向智慧化，因此刀具磨耗預測與換刀時間點顯得更加重要。然而，比起常見的車、銑加工製程，與內螺紋攻牙有關的研究相對較少。攻牙是一項複雜的製程，相比於其他製程更為不穩定，容易伴隨著卡屑、斷刀及螺紋不合格等問題。此外，攻牙製程常為加工的最後幾道工序，若發生問題時要補救相當耗時費力，嚴重時甚至需要將整個工件報廢，造成巨大的損失。回顧文獻中對於攻牙的研究，大多是使用動力計直接測得攻牙時之扭矩值作為參考依據，但動力計價格昂貴、剛性不佳，不易在產線上被使用，因此本研究在不外加感測器的條件下，直接從工具機控制器擷取控制馬達的扭矩指令訊號，並利用卷積神經網路進行模型建立與刀具狀態分類。卷積神經網路在圖像和聲音的辨識上有相當優秀的成果，再加上近年來硬體設備的進步與演算法的優化，使得運算速度大幅提升。
本研究利用不同方式將控制器所擷取到之攻牙扭矩指令訊號轉換成圖片後，輸入卷積神經網路中進行模型訓練，再使用此模型將圖片分類成正常、嚴重磨耗及螺紋不合格三種類別，以利找出最佳的換刀時間點，並減少檢測螺紋的時間。當訊號被判定為刀具過度磨耗或崩刃時，則代表需要更換刀具；若訊號被判定為螺紋品質不合格時，便僅需針對該螺絲孔做出相對應之處理，不必再以螺紋塞規逐一檢測螺紋。本研究搭配卷積神經網路來判斷換刀時機與減少螺紋檢測時間，確實可增加刀具使用率和降低產線上所需的人力成本，具工業應用高度潛力。

關鍵字

卷積神經網路 ； 攻牙刀具狀態 ； 換刀時間判斷 ； 主軸扭矩指令 ； 螺紋品質檢測

並列摘要

Recently, Industry 4.0 has become a trend for manufacturing industries, and under the core concept of the machine tools industry will be gradually evolved from pure automation to smart factories. The key issue resides in the monitoring of cutting tools conditions; consequently, the timing of the cutting tools changes. However, literatures on tapping are limited. Tapping is quite a complicated process, which is an unstable process and often accompanies with chip clogging, tool breakage, bad thread quality, and other failures. Moreover, since tapping is near the finishing process, the associated repair work is very time-consuming and laborious when problems should occur. If workpiece cannot be repaired, previous efforts may come in vain. Previous research on tapping shows torque signals were collected by dynamometers. However, dynamometers are expensive and not practical to be applied in production environments. In this study, torque command signals directly acquired from CNC controller are applied with a use of convolutional neural network. The model established is shown to distinguish healthy tool from the worn tool.
In the study, various approaches were used to convert torque command signals to images, which were then imported in convolutional neural network for model training. Convolutional neural network is particularly powerful in recognizing images and voices. In particular, the improvement of hardware and algorithms dramatically speeds up computing process.
By means of the above approach, best timing to change taps can be determined and the time required to check threads can be drastically reduced.

並列關鍵字

convolutional neural network ； tapping tool condition ； timing of changing tools ； spindle torque command ； thread quality inspection

參考文獻

[1] D. H. Hubel and T. N. Wiesel, "Receptive Fields, Binocular Interaction And Functional Architecture In The Cat's Visual Cortex," Journal of Physiology, p. 106–154, 1962.

[2] F. Kunihiko, "Neocognitron: A Self-organizing Neural Network Model for a Mechanism of Pattern Recognition Unaffected by Shift in Position," Biological Cybernetics, vol. 36, pp. 193-202, 1980.

[3] Y. LeCun, L. Bottou, Y. Bengio and P. , "Gradient-Based Learning Applied to Document Recognition," Proceedings of the IEEE, vol. 86, Nov 1998.

[4] A. Krizhevsky, I. Sutskever and G. E. Hinton, "ImageNet Classification with Deep Convolutional Neural Networks," Advances in Neural Information Processing Systems 25, 2012.

[5] A. A. D. Sarhan, M. Sayuti and . M. Hamdi, "Reduction of power and lubricant oil consumption in milling process using a new SiO2 nanolubrication system," The International Journal of Advanced Manufacturing Technology, 2012.