Name
Abstract | ||
|---|---|---|
Predicting remaining useful life (RUL) of plasma equipment becomes an important issue for semiconductor manufacturing in this decade. If RUL can be accurately estimated, the schedule of maintenance can be proper to moderate the waste and cost of the production. Digital Radio Frequency Matching Box (RF-MB) is an essential equipment in the semiconductor manufacturing process. The status of RF-MB will be recorded by the Fault Detection and Classification (FDC). In order to establish the RUL of RF-MB, we use Fisher Discriminant Analysis (FDA) for feature selection to concentrate the leading variables in FDC. We marked the first 2 days of the RF-MB operation as "Good" and marked the last 2 days before the failure of RF-MB as "Bad". We used extended Classifier System with continuous-valued inputs (XCSR) to learn the well-labeled FDC data. The results show that XCSR can quickly find patterns and meaningful variables. The average accuracy of XCSR is 97.3% and the average missing rate of rules is only about 1.6%. The results confirmed that XCSR is capable of alerting related operator before the plasma component reaching its residual life. In the future, we will use XCS with Function approximation (XCSF) to more accurately approximate the function of RUL. We look forward to building a complete assessment of RUL.
|
Year | DOI | Venue |
|---|---|---|
2019 | 10.1145/3319619.3326879 | GECCO |
Keywords | Field | DocType |
Fisher Discriminant Analysis (FDA), eXtended Classifier System (XCS), Digital Radio Frequency Matching Box (RF-MB), Remaining Useful Life (RUL) | Computer science,Artificial intelligence,Machine learning | Conference |
ISBN | Citations | PageRank |
978-1-4503-6748-6 | 0 | 0.34 |
References | Authors | |
0 | 5 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Liang-Yu Chen | 1 | 0 | 1.35 |
| Jia-Hua Lee | 2 | 0 | 0.68 |
| Ya-Liang Yang | 3 | 0 | 0.34 |
| Ming-Tsung Yeh | 4 | 0 | 0.34 |
| Tzu-Chien Hsiao | 5 | 42 | 11.74 |