The Development of Tool Condition Monitoring in the Milling of Inconel 718 by Recurrent HMMs

• Main Authors: Chung-Ying Wang, 王崇穎
• Other Authors: Ming-Chyuan Lu
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/738fdq

碩士 === 國立中興大學 === 機械工程學系所 === 106 === The Inconel-718 is the key material for the aero industry, but the serious tool wear always make it a challenge to improve the quality and machining efficiency. The tool condition monitoring provides a solution to optimize the tool exchange time and machining efficiency. However, the uncertainty of system and material variation make the system not easy to implement the developed system in the laboratory to the industrial site. Therefore, how to improve the robustness of the system plays an important role to implement the tool wear monitoring system in the production line. In this research, a modified Recurrent Hidden Markov Model (RHMMs) is proposed to improve the robustness of the monitoring system. In the system development and evaluation, the variation of system is designed by changing the clamping force on the vise and the diameter of cutting tool. Finally, multi-sensor feature and decision fusion was adopted to improve the overall performance of the monitoring system. In experiments, the vibration signals in the X and Y directions were collected simultaneously during the machining of the straight line, along with the Acoustic Emission (AE) signals. The results show that the classification rate of tool wear monitoring with the constant clamping force in machining is improved from 85% to 94.8% with the implement of the modified RHMMs comparing to the one with traditional HMMs. With the varying clamping force setup, the classification rate is improved from 86.39% to 90.83%. In the analysis of applying the feature and decision fusion to the BPNN, HMMs, and RHMMs classifiers, the improvement of system robustness is also observed for the system with the RHMMs classifier. Although no clear difference of the classification rate between applying three classifiers is observed by evaluating the system by signals with the constant clamping force, the classification rate could be improved from 92.78% and 94.4% to 97.78%, respectively, by adopting the feature fusion integrated with RHMMs classifier comparing to the integration with BPNN and HMMS classifiers. With the implementation of decision fusion integrated with RHMMs, the classification rate could be improved from 93.96% and 90.83% to 96.39% comparing to the BPNN and HMMs classifiers.