(92d) Industrial Implementation of on-Line Multivariate Quality Control (Part II: Long-Term Performance, Model Maintenance, and Model Leveraging)

Implementation of on-line multivariate quality control was discussed previously (Chiang and Colegrove, 2005). Robust principal component analysis (PCA) was applied to historical quality data to remove outliers. The remaining data, representing normal process variation, is used to build a PCA model. One advantage of multivariate analysis is that PCA can detect a change in variable correlation, which is undetectable or not easily detectable using univariate control charts. The T2 statistic is used for fault detection and the contribution chart is used for fault identification. The model has been implemented on-line and proven to be effective in detecting and identifying abnormal product lots.

The concept of multivariate quality control is the same as multivariate statistical process control (MSPC). Both methods focus on fault detection and identification. The difference is that multivariate quality control focuses on quality data while MSPC focuses on process data. This talk discusses learning experiences from the on-line implementation. In particular:

- long-term performance: Monitoring of the T2 statistic provides reasonably immediate evidence of the success or failure of newly implemented process improvements

- model maintenance: Model updates can easily be performed to reflect new production targets.

- model leveraging: Models can, in some situations, be used to monitor a similar product in a different facility, thus reducing development time.

Reference:

L. Chiang and L. Colegrove, Industrial implementation of on-line multivariate quality control, AIChE Annual Meeting, Cincinnati, OH, 2005. Paper 320a.