Converting Data into Information - The Application of Multivariate Statistical Techniques...
Converting Data into Information – The Application of Multivariate Statistical Techniques in Agrochemicals Production
Alan Brown, PhD, Principal Technical Expert – Syngenta
Syngenta is a world-leading agribusiness providing crop protection products (herbicides, insecticides and fungicides), plants and seeds for farmers and growers. Manufacture of the chemical active ingredients is undertaken at a number of sites using a combination of batch and continuous processing. A characteristic of these sites is the production of large amounts of data. This presentation illustrates the use of the modeling and multivariate capabilities within JMP Pro to turn such data into information that enables improved process understanding. Three examples of increasing complexity are presented. The first utilizes principal components analysis (PCA) of six variables to investigate the reason for performance differences between two production units. The second considers 34 variables (for example, flows, temperatures, pressures, etc.) from a continuous process and the use of classification techniques, such as recursive partitioning, boosted trees, PLS-DA and neural networks, to predict the polymorphic form of a product. The same set of input variables is used to model the amount of a group of related chemical impurities using PLS and neural networks. The third example uses PCA/PLS and recursive partitioning to examine the differences between three parallel production streams. Data on 14 process variables were collected for 30 batches, 10 for each stream. Approximately 400 observations, at one-minute time intervals, were captured per variable per batch.
Converting Data into Information – The Application of Multivariate Statistical Techniques in Agrochemicals Production
Alan Brown, PhD, Principal Technical Expert – Syngenta
Syngenta is a world-leading agribusiness providing crop protection products (herbicides, insecticides and fungicides), plants and seeds for farmers and growers. Manufacture of the chemical active ingredients is undertaken at a number of sites using a combination of batch and continuous processing. A characteristic of these sites is the production of large amounts of data. This presentation illustrates the use of the modeling and multivariate capabilities within JMP Pro to turn such data into information that enables improved process understanding. Three examples of increasing complexity are presented. The first utilizes principal components analysis (PCA) of six variables to investigate the reason for performance differences between two production units. The second considers 34 variables (for example, flows, temperatures, pressures, etc.) from a continuous process and the use of classification techniques, such as recursive partitioning, boosted trees, PLS-DA and neural networks, to predict the polymorphic form of a product. The same set of input variables is used to model the amount of a group of related chemical impurities using PLS and neural networks. The third example uses PCA/PLS and recursive partitioning to examine the differences between three parallel production streams. Data on 14 process variables were collected for 30 batches, 10 for each stream. Approximately 400 observations, at one-minute time intervals, were captured per variable per batch.