Identification of shift variation in bivariate process using pattern recognition technique
In quality control, the identification of unnatural variation in mean shifts is a challenge when dealing with two correlated quality characteristics (bivariate). Various schemes based on statistical process control pattern recognition (SPCPR) approach have been proposed to monitor the presence of un...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/560/1/24p%20MOHAMAD%20AZRUL%20AZHAD%20MOHD%20HAIZAN.pdf http://eprints.uthm.edu.my/560/2/MOHAMAD%20AZRUL%20AZHAD%20MOHD%20HAIZAN%20COPYRIGHT%20DECLARATION.pdf http://eprints.uthm.edu.my/560/3/MOHAMAD%20AZRUL%20AZHAD%20MOHD%20HAIZAN%20WATERMARK.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In quality control, the identification of unnatural variation in mean shifts is a challenge when dealing with two correlated quality characteristics (bivariate). Various schemes based on statistical process control pattern recognition (SPCPR) approach have been proposed to monitor the presence of unnatural variation and diagnose its sources. However, existing studies have focused either on limited sources of unnatural variation and/or lack in avoiding false alarm. In terms for unnatural variation, the classification of reciprocating patterns such as upward-downward and downward-upward shifts was less reported. In order to enhance its capability, this study aims at proposing an improved design of SPCPR scheme for enabling classification of nine bivariate patterns with high accuracy and reduced false alarm. There were two schemes investigated in this study: (i) Statistical Features-Multilayer Perceptron (SF-MLP) and (ii) Integrated Multivariate Exponentially Weighted Moving Average-Multilayer Perceptron (MEWMA-MLP). Input representation for the MLP recogniser in both schemes was designed based on summary statistical features and it was systematically selected using the design of experiments analysis. The average run lengths (ARL0, ARL1) and recognition accuracy percentage (RA) were used as the performance measures. In monitoring performance, the MEWMA-MLP provided a longer run of ARL0 (386.8 ~ 606.9) and ARL1 (2.76 ~ 9.63) compared to the SF-MLP (ARL0 = 160.8 ~529.3, ARL1 = 2.25 ~ 8.91). This result suggests that the MEWMA-MLP is better in avoiding false alarm, which exceeded the de facto level (ARL0 = 370) but requires a slightly longer run to detect the unnatural variation. In diagnosis performance, the MEWMA-MLP gave an improved range of recognition accuracy (RA= 86.5 ~ 99.3 %) compared to the SF-MLP (RA= 83.3 ~ 97.7 %) in classifying the sources of unnatural variation. Overall, this study opens a new perspective to enhance the capability of SPCPR scheme for the application of bivariate quality control. |
|---|