nonlinearity is removed by subtracting the mean trajectories from reference set trajectories prior to analysis. Another statistic that is calculated from MPLS model is the Variable Influence on Projection (VIP) to investigate the effects of important process variables (predictors) on quality variables (predictees). As the formulation and interpretation details provided in Section 6.4.2, process variables that have contributions larger than 1 can be considered to exert more influence on quality as far as MPLS projection is concerned. Figure 6.48 summarizes the mean values of VIP set, i.e., over the entire course of batch run and according to these plots variables 5, 7, 8, 9, 13 and 14 are found to be important, which is meaningful since these variables carry physiological information and hence are expected to be effective on the quality.

Process monitoring and quality prediction stage: Developed MPLS model is used to monitor finished batches to classify them as 'good' or 'poor' based on how well they follow similar trajectories to achieve 'good' quality product. The same fault scenario with a small downward drift on glucose feed (see Figure 6.44 and Table 6.8) in MPCA based monitoring is used to illustrate end-of-batch MPLS framework. MPLS model is also used to predict product quality as soon as the batch finishes providing information ahead

Table 6.11. Cumulative percent variance captured by MPLS model on each quality variable

LV no.

Was this article helpful?

0 0

Post a comment