應用小波分析與類神經網路於管制圖非隨機樣式之偵測

Control Chart Patterns Recognition using Wavelet Analysis and Neural Control Chart Patterns Recognition using Wavelet Analysis and Neural Network

羅完元(Tzeng-Yuan Lo)

指導教授 ： 鄭春生

元智大學/工程學院/工業工程與管理學系/碩士(2005年)

摘要

摘要

在統計製程管制中，當製程存在可歸屬原因時，管制圖上會呈現特定的非隨機性樣式的變化。正確辨識非隨機性樣式，可以縮小診斷製程異常原因之範圍而有助於規劃改善方案。本研究目的是發展一個可有效辨識趨勢、週期性及平均值偏移等非隨機性樣式種類及正常數據之辨識系統。

本研究提出一種以類神經網路為基礎之辨識程序，做為管制圖非隨機性樣式之辨識系統。在網路輸入向量的設定上分別以 (1) 原始數據及 (2) 結合小波理論之數據前處理過程所得的特徵值兩種方法進行實驗比較與探討。本文的主要貢獻為小波理論中之多重解析度分析過程所得之各階特徵係數因具有良好的能量集中性質，而有助於以較小的網路輸入向量描述各樣式的特徵及其差異性，減少執行運算及訓練的時間。

本研究所建立之辨識系統是以正確辨識率來評估其效益。經模擬法分析證明本研究所提出之小波特徵擷取方法確實可提高應用時之辨識能力，同時亦降低原有網路結構的複雜性。

關鍵字

非隨機性樣式 ； 類神經網路 ； 小波理論 ； 多重解析度分析

並列摘要

ABSTRACT

In the statistical process control, control chart patterns recognition plays an important role. When the process existed may belong to the assignable causes, a control chart can present the specific non-random patterns. Correctly recognizes the non-random patterns may reduce examines scope of the regulation abnormal reason to be helpful to the improvement plan. The purpose of this research is to develop a system to recognize the trend, cycle, shift non-random patterns and the normal data effectively.

In this research, we proposed a neural network-based pattern recognizer for the analysis of control chart patterns. In addition, we have developed and compared with two kind of input vectors: raw data and features that extrascted from wavelet theory and multi-resolution analysis. The concepts of wavelet theory and multi-resolution analysis (MRA) have be used for data pre-processing. It seems more capable of detecting unnatural patterns as well as describing the key features of the specific pattern detected and the excuting time of artificial neural network.

Since the pattern recognizer of this research is appraised its benefit by the rate of correct classification. The results of simulation have show that the pattern recognizer using features based on wavelet theory and multi-resolution analysis can recognize the control chart patterns better than raw data and the structure of neural network is simply, too.

並列關鍵字

non-random patterns ； artificial neural network ； wavelet theory ； multi-resolution analysis

參考文獻

1. Aradhye, H. B., B. R. Bakshi, R. A. Strauss, and J. F. Davis, “Multisacle SPC using wavelets-theoretical analysis and properties,” AICHE Journal, 49, 939-958 (2003).

2. Crowder, S. V. and Hamilton, M. D., “An EWMA for monitoring a process standard deviation,” Journal of Quality Technology, 24, 12-21 (1992).

3. Cheng, C. S., “A neural network approach for the analysis of control chart patterns,” International Journal of Production Research, 35, 667-697 (1997).

4. Cheng, C. S., and Hubele N. F., “A pattern recognition algorithm for an control chart,” IIE Transactions, 28, 215-224 (1996).

5. Donoho, D. L., and Johnston, I. M., “Ideal spatial adaptation via wavelet shrinkage, “Biometrika, 81, 425-455 (1994).