Process upsets and product quality issues often result in loss of asset capability or customer concerns. When these unfavorable conditions happen, it is important to effectively and promptly identify the root cause(s) to recover to normal operating conditions.

Multivariate statistical analysis has been proven to be effective in process troubleshooting for The Dow Chemical Company. This talk will discuss the use of regression techniques such as partial least squares PLS, classification techniques such as discriminant partial least squares (DPLS), and dimensionality reduction techniques such as principal component analysis (PCA) to diagnose various complex manufacturing problems at Dow. Practical aspects of industrial data treatment such as variable selection, outlier detection, missing value estimation will be discussed as well. Industrial examples will be given to illustrate how to make data-based decision making for process troubleshooting.

Extended Abstract Status: Not Uploaded