| Summary: | 碩士 === 國立雲林科技大學 === 工業工程與管理系 === 107 === Recognizing unnatural control chart patterns is an important issue. Many scholars used the most common method which is the neural network to recognize control patterns. The neural network has supervised learning network and unsupervised learning network. Both of them have excellent performance. Most of the studies assume that the process data are mutually independent, but autocorrelation process data are common in continuous processes. This study uses Support Vector Machines and Self-organizing Map to recognize control chart patterns in the autocorrelation process, and explore the performance of supervised learning network and unsupervised learning network in recognizing patterns.
The proposed method uses wavelet analysis for data preprocessing to extract eigenvalues and reduce signal noise, and then uses the results as input terms of support vector machines and self-organizing image network graph to improve the recognition effect. The results show that the average recognition accuracy without wavelet analysis is more than 80%. The average recognition accuracy of wavelet analysis is over 94%. Thus, wavelet analysis can effectively improve the recognition accuracy.
|
|---|
