D.1 Statistical process monitoring and optimisation

Readily available online measurement data can be effectively used in industrial practice for monitoring process performance. The statistical tools enables unveiling multivariate behaviours and thus to map the performance down into a space of comparatively low dimensionality where supervision is greatly facilitated. In this project research aspects of monitoring continuous and semi-continuous processes are considered.
On continuous processes the project deals with multivariate monitoring of multi-modal processes based on principal component analysis (PCA). For continuous chemical process plants operating at one operation mode this technique has been proven to work well. The sound statistical basis of the technique is however lost when the monitored process has more than one operation mode. The result is a reduction is the fault detection sensitivity.
The purpose of the project is to develop approaches to enhance the monitoring capability of the technique when monitoring processes with more than one operation mode. It is intended to do this by using localised PCA representations, either based on a number of pre-determined representations or by constant updates of the representation using exponential weighted moving average. The analysis will be performed on data from the energy-integrated distillation column located at CAPEC, DTU.
Fed-batch fermentation processes are time consuming and difficult to model using first principle models and these models are difficult to apply in production. Instead the process can be modelled by using data based modelling. The fault diagnosis utilises a statistical model based on either PCA (principal components analysis) or PLS (projection to latent structures) to describe normal behaviour. Deviations from normal conditions can easily be detected and displayed to the process operators, which leaves them time to take action to eliminate the fault. The methodology has been demonstrated on a fermentation process.