Statistical process control in stamping process using pattern recognition approach
Unnatural process variation (UPV) is a vital quality problem in the metal-stamping process that contributes to the causes of poor product quality. In the production site, sources of UPV are usually found based on special causes. Recently, there are still debates amongst researchers to find an effect...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English English |
| Published: |
2018
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/481/1/24p%20NORASULAINI%20ABDUL%20RAHMAN.pdf http://eprints.uthm.edu.my/481/2/NORASULAINI%20ABDUL%20RAHMAN%20COPYRIGHT%20DECLARATION.pdf http://eprints.uthm.edu.my/481/3/NORASULAINI%20ABDUL%20RAHMAN%20WATERMARK.pdf http://eprints.uthm.edu.my/481/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Unnatural process variation (UPV) is a vital quality problem in the metal-stamping process that contributes to the causes of poor product quality. In the production site, sources of UPV are usually found based on special causes. Recently, there are still debates amongst researchers to find an effective technique for online monitoring and diagnosing UPV sources. Control chart pattern recognition (CCPR) is the most investigated technique. The existing CCPR framework was mainly developed by using raw data-based artificial neural network (ANN) recognisers, whereby the recognisers were trained by using artificially generated statistical process control (SPC) chart patterns. This is because the real SPC samples were commonly confidential, not economically available or difficult to obtain in a large amount for sufficient training. In this study, a new CCPR framework by using statistical features ANN recogniser with real SPC samples as the training control chart patterns was investigated. The SPC data were directly taken from the metal-stamping process. This framework is able to monitor and diagnose several root cause errors, such as dimension out, double punch, high burr and rivet slanting with a short recognition window size (10 subgroup samples). Based on dynamic data training, the proposed recogniser resulted in a better recognition accuracy (normal pattern = 100%, unnatural pattern = 100%) as compared to the raw data-based ANN (normal pattern = 66.67%, unnatural pattern = 26.92%). To ensure its applicability for real manufacturing process, the performance of the framework was successfully validated by using new SPC samples. This finding would be useful to avoid erroneous interpretation of root cause for defects amongst quality practitioners. |
|---|