| Summary: | 碩士 === 國立中興大學 === 機械工程學系所 === 98 === As the demand of the small feature and high accuracy for optical, electronic, and biomedical devices continuously increases, the micro mechanical machining plays an important role for improving their manufacturing quality and efficiency. Due to the higher tool wear rate than conventional counterpart, the tool wear monitoring in the micro machining draws much more attention than before.
The objective of this thesis is to analyze the performance of tool wear monitoring system integrated with the spindle vibration and acoustic emission signal obtained from the spindle housing, as well as the study of the effect of system parameters on the system performance. For improving the classification rate, a decision fusion algorithm was also adopted to integrate the decision made from the spindle vibration and AE signal for tool wear monitoring.
A micro tool condition monitoring system integrated by sensor system, signal transformation, feature selection, and classifier was developed in this study. In which, the FFT transformation was used for transforming the time domain signal to the frequency domains, the class mean scatter criteria was used to select the features closely related to the tool wear condition, and the Fisher linear discriminant function was the basis for designing the classifier. In the analysis of the parameters effect on the system performance, the bandwidth sizes of frequency domain signal, the number of the selected features and the change of contact between the tool and tool holder were studied.
The results show that the AE and vibration signal obtained from the spindle housing can be used to detect the change of tool wear on a micro end mill. The energy of the vibration signal was observed to increase as the tool wear proceeds. However, when the tool was reinstalled between each pass of cutting, the same trend will not be kept. For the AE signal, the energy of signal between 60 kHz to 70 kHz was observed to increase as tool wear proceeds. In addition, the energy of signal between 330 kHz to 400 kHz was observed to change as the tool was reinstalled between each pass of cutting. In the tool wear classification results, the effect of system parameters such as bandwidth size of frequency domain signal, and the number of features selected on the system varied for the different direction of vibration signal. The best parameters selected for the case with the X direction vibration signal was two features selection along with bandwidth size of 235Hz, but two or three features selection along with bandwidth size of 120Hz was found to be best for the Y and Z direction vibration signal. Moreover, by modifying the Fisher linear discriminant function, the increase of feature number can improve the classification rate by 5% to 10%. In consideration of the AE signal case, the increase of the bandwidth size was observed to improve the classification rate to 75% with bandwidth size of 53.3 kHz. Finally, by integrating the decision from the discussed four signals, the classification can be improved and reaches 95%.
|
|---|
