The availability of data is an important success factor for any modelling and simulation project. Nowadays mills have access to extensive process data. While there have been high levels of investment to install and operate databases, data management software, and process monitoring systems, the value currently derived from these systems tends to be low. This puts pressure on the promoters of the investment and their common belief that this high amount of data will be a value in itself. In many cases this proves to be wrong. Typical reasons besides a lack of expertise to properly analyse data are that parts of the process are not available to the database system or that the process is not equipped with some perceived need to solve a specific technological question. The ‘old-fashioned’ sampling technique is still the best practise, especially when dealing with complex wet end problems and when setting up mill balances for pulp, water, energy, and detrimental substances (Hutter and Kappen, 2004)
12.2.1 Methods for the Analysis of Data
Process monitoring systems facilitate the study of historical data and establish trends. However, the possibilities of advanced data treatment for process control and optimisation are not always well known or exploited by papermakers. The first step is to classify the data and the second step is to analyse them in order to answer a specific question. Depending on the final aim, different analysis techniques can be applied, methods being quite often complementary:
1. Classic statistical analysis techniques, time series analysis, experimental design, D-Optimal design, and trial-and-error methods are often used for quick and adequate solution of isolated problems (Box, Hunter, and Hunter, 1978).
2. Building physicochemical deterministic models is the optimum approach when the overall behaviour of the process or one of its sections must be modelled, in order to, for instance, design new equipment or study the kinetics of flocculation (Blanco et al., 2002; Negro et al., 2005; Thomas, Judd, and Fawcett, 1999).
3. The third approach is a combination of approaches (1) and (2) above. Sometimes the physics and/or chemistry of the phenomenon are not known, but a general model has to be built to predict process evolution, optimise process variables, or simulate scenarios. Advanced data analysis tools—e. g., multivariate analysis techniques and artificial neural networks—can provide models with better performance than techniques from approach 1, if the behaviour of the process, section, or equipment to model is complex and unknown (Miyanishi and Shimada, 1998; Masmoudi, 1999; Parrilla and Castellanos, 2003, Blanco et al., 2005).