Latent variable based key process variable identification and process monitoring for forging

Rapid developments in sensing technology have significantly increased the accessibility of processing condition information. Yet product quality may be primarily affected by a few critical process variables. Identifying key process variables will aid efficient process information collection and focused process monitoring. In this paper, key process variables are identified through a two-step procedure, where globally important process variables are identified on the basis of overall quality variables. Localized identification of key process variables for specific quality characteristics is performed using the regression coefficient matrix and direct clustering algorithm. On the basis of the latent variable modeling approach, in-line process monitoring is performed as well as prediction of the quality features. Real data sets collected from a crankshaft forging are used to evaluate the performance.