Home > Page

Applications within process analysis


A great number of applications within process analysis have been carried out using multivariate analysis. Data compression, variable reduction, indirect measuring, prediction, and classification, are some of the tasks that have been solved by projection methods. A few examples:

  • Important process variables in crystallization processes are identified by multivariate analysis in combination with experimental design. This knowledge is used to redesign the process and to improve the process control.
  • Raw material control is improved by replacing one-dimensional specifications with multivariate specifications, because it is often the combination of quality variations in several parameters that determine the total quality.
  • Multivariate data analysis from a large set of process measurements showed which sensors that gave the same or little information about important quality parameters. Thus redundant sensors could be dropped, instrumentation simplified, and maintenance costs reduced.
  • Expensive, inaccurate, and infrequent dioxin emission measurements in steel works have successfully been replaced by indirect measurements of chlorinated benzenes.
  • Octane numbers in gasoline samples are predicted on-line from Near Infra Red spectroscopic measurements.
  • Paper quality has been related to process measurements and raw material composition. This is used for production planning and to identify causes to bad quality.
  • The feedback to the control system for distillation columns is speeded up by replacing chromatographic measurements of the output composition by temperature profile measurements from several sensors placed on the column.
  • Multivariate process monitoring may give fewer false "alarms" and earlier warnings of drift and real problems. This is because univariate control limits are often too narrow to handle normal process variation (noise), or too wide to detect drift early.