Online batch-end quality estimation: does laziness pay off

The influence of measurement noise on PLS-based batch-end quality prediction*

“…The advantage of TSR is that it only requires a single PLS model to predict the final batch quality at every sample instance throughout the batch instead of K different models. Previous research by the authors has shown that it shows similar performance to the training of a new PLS model for every time at which a prediction is needed (Gins et al, 2009). TSR uses a regression model to estimate the final scores T new,k of a new batch based on the trimmed scores T * new,k .…”

Discriminating between critical and non-critical disturbances in (bio-)chemical batch processes using multi-model fault detection and end-quality prediction

“…Of the missing data methods investigated by García-Munoz et al, trimmed scores regression (TSR , ) was illustrated to exhibit superior performance. Additionally, previous research by the authors shows that the performance of TSR for batch-end quality prediction is comparable to (but less labor-intensive than) training a different MPCA or MPLS model for each time point at which a score or quality prediction is requested (i.e., K different models if a quality prediction must be made at each time instant) . Therefore, TSR is used in this study for obtaining online MPCA and MPLS scores and MPLS quality estimates.…”

“…Additionally, previous research by the authors shows that the performance of TSR for batch-end quality prediction is comparable to (but less labor-intensive than) training a different MPCA or MPLS model for each time point at which a score or quality prediction is requested (i.e., K different models if a quality prediction must be made at each time instant). 26 Therefore, TSR is used in this study for obtaining online MPCA and MPLS scores and MPLS quality estimates. Finally, a major advantage of TSR is that a single model can be used at all times of the batch operation, minimizing the effort required for model identification.…”

Extending discrete batch-end quality optimization to online implementation1

“…This is an order of magnitude larger than the crossvalidation MSE values observed for the single feed rate changes in the first study. Therefore, model performance is improved via variable selection [Gins et al, 2009]. The selection procedure results in the retention of 6 input variables: time, reactor volume, feed rate, cooling water flow, base flow, and acid flow.…”

“…More recently, Ündey et al [2003] also used PLS to predict the final penicillin concentration for the simulated fermentation of Birol et al [2002]. The authors also reported successful (online) batch-end quality prediction for an industrial polymerization process [Gins, 2007, Gins et al, 2009. Baffi et al [2000] used nonlinear modelling for a benchmark pH neutralization system.…”