4.11 Multivariate Statistical Process Control and Process Control, Using Latent Variables ☆

Databases containing process data collected in industry are becoming increasingly large. Practitioners frequently turn to these data to gain process understanding with the aim to improve the process, to establish procedures for process monitoring, and if possible to establish procedures for some form of process control in real time. The types of data required to address each one of these objectives as well as the types of models and procedures that are suitable for such purposes depend on the objective. Utilizing databases for such activities should be planned carefully to obtain scientifically viable results. In the industrial database, some of the data may have been collected by applying some form of design of experiments (DOEs) but the majority of them have been collected during routine production. As a result, most of the data are noncausal in nature. They consist of highly correlated variables with many missing measurements and low content of information in any one variable, due to the low signal-to-noise ratios. Latent variable methods can deal with such data and as a result they have received a lot of attention by industrial practitioners for troubleshooting, process understanding and improvement multivariate monitoring and with the appropriate amount of Designed Experiments for control schemes and product transfer.