Performance of OSS and NICN Strategies for Batch Process Monitoring

ALVAREZ RODRIGO; BRANDOLIN ADRIANA; SANCHEZ MABEL

Nashville, TN-USA

Congreso; 2009 AIChE Annual Meeting; 2009

American Institute of Chemical Engineers

During the last years, several works dealing with statistical process monitoring and controlhave been presented. In the particular case of batch processes, detection and identification related tasks, rather than root cause isolation and diagnosis, have been the main concern of researchers working on the area. Most popular strategies are based on latent variable projection techniques, such as Multiway Principal Component Analysis (MPCA), Multiway Independent Component Analysis (MICA) or Multiway Partial Least Squares (MPLS).Mason et al. (1997) proposed a different approach to carry out detection and identification tasks in the space defined by the original measured variables (i.e. without performing any variable projection), using only one statistic. The Hotelling´s statistic (T2-statistic) is first monitored in order to establish whether the process is on or out of control. If an out of control situation is declared, the T2-statistic is decomposed into the contributions of each measured variable, and these contributions are further inspected to identify the suspicious variables.Due to the combinatorial nature of the formulation proposed by Mason et al. (1997), severalpossible decompositions are obtained, increasing the complexity of the identification procedure. A straightforward method to decompose the T2-statistic as a unique sum of variable contributions, named Original Space Strategy (OSS) was developed by Alvarez et al. (2007). This method also provides an explanation about the physical meaning of the negative contributions, which are also commonly found in classically used contribution plots (Westerhuis et al., 2000), and estimates a bound for them.Recently, Alvarez et al. (2009) presented a new strategy to estimate the influence of a givenvariable on the T2-statistic value. In this approach, the contribution of each variable is measured in terms of the distance between the current observation and its Nearest In Control Neighbour (NICN), which is evaluated solving an NLP optimization model.The aim of this work is to present a comparative analysis of the performances of the NICNapproach and the OSS strategy. A benchmark fed-batch penicillin fermentation process (due to Birol et al. (2002)) is used as case study to evaluate the techniques´ performances.