Design optimization for the two-stage bivariate pattern recognition scheme
In manufacturing operations, unnatural process variation has become a major contributor to a poor quality product. Therefore, monitoring and diagnosis of variation is critical in quality control. Monitoring refers to the identification of process condition either it is running within in statisticall...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English English |
| Published: |
2015
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/1380/2/MOHD%20SHUKRI%20MOKHTAR%20COPYRIGHT%20DECLARATION.pdf http://eprints.uthm.edu.my/1380/1/24p%20MOHD%20SHUKRI%20MOKHTAR.pdf http://eprints.uthm.edu.my/1380/3/MOHD%20SHUKRI%20MOKHTAR%20WATERMARK.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In manufacturing operations, unnatural process variation has become a major contributor to a poor quality product. Therefore, monitoring and diagnosis of variation is critical in quality control. Monitoring refers to the identification of process condition either it is running within in statistically in-control or out-of-control, whereas diagnosis refers to the identification of the source of out-of-control process. Selection of SPC scheme becomes more challenging when involving two correlated variables, which are known as bivariate quality control (BQC). Generally, the traditional SPC charting schemes were known to be effective in monitoring aspects, but there were unable to provide information towards diagnosis. In order to overcome this issue, many researches proposed an artificial neural network (ANN) - based pattern recognition schemes. Such schemes were mainly utilize raw data as input representation into an ANN recognizer, which resulted in limited performance. In this research, an integrated MEWMA-ANN scheme was investigated. The optimal design parameters for the MEWMA control chart have been studied. The study focused on BQC with variation in mean shifts (μ = ±0.75 ~ 3.00) standard deviations and cross correlation function (ρ = 0.1 ~ 0.9). The monitoring and diagnosis performances were evaluated based on the average run length (ARL0, ARL1) and recognition accuracy (RA) respectively. The selected optimal design parameters with λ=0.10, H=8.64 gave better performance among the other designs, namely, average run length, ARL1=3.24 ~ 16.93 (for out-of-control process) and recognition accuracy, RA=89.05 ~ 97.73%. For in-control process, design parameters with λ=0.40, H=10.31 parameter gave superior performance with ARL0 = 676.81 ~ 921.71, which is more effective in avoiding false alarm with any correlation. |
|---|