Inferential properties, also known as "soft sensors" or "virtual analysers" can be extremely valuable. Even if an on-stream analyser is installed, an inferential technique will usually respond much more quickly to process changes and can therefore be used to significantly reduce disturbances to product composition. In many cases inferentials can be sufficiently accurate to obviate the need for an on-stream measurement.
The package is not a substitute for sound process engineering. It is intended to provide a user-friendly tool to permit rapid analysis of historically collected process data. Interpretation of the results and selection of the most appropriate calculation still requires a good understanding of the process operation.
A wide range of linear and non-linear functions, involving few or many process measurements can quickly be evaluated. Analysis statistics and graphical functions allow the user to easily identify suspect measurements and remove them. Time dependent performance statistics can be used to identify any systematic changes in accuracy and permit laboratory update functions to be optimised.
Correlations can be saved to disk. This permits the user to explore the robustness of the correlation. For example, data could be split into a 'training' set used to develop the correlation and a 'testing' set to check its reliability. Similarly a correlation developed some time ago may be applied to more recently collected data to determine whether its accuracy has degraded over time.
Next (Parallel Coordinates)
Back to Toolkit