GB2438028A

GB2438028A - Multivariate Visualisation of a Batch Process

Info

Publication number: GB2438028A
Application number: GB0609443A
Authority: GB
Inventors: Alan James Mason
Original assignee: Individual
Current assignee: Individual
Priority date: 2006-05-12
Filing date: 2006-05-12
Publication date: 2007-11-14
Also published as: GB0609443D0

Abstract

Quality and process variables of a batch process are displayed using parallel axes (A01 to A13). The displayed values each relate to a single batch (B1 to B5). Batches can be selected and displayed with differentiation, allowing the relationship of the variables for the specific batch to be visualised (B3). Where a process variable for a batch is displayed as multiple values these values may be plotted against time and the resultant curves for multiple batches overlaid in a single graph. Display differentiation for curves relating to selected batches can be applied. The parallel axis graph and the multiple value curve graph can be simultaneously displayed and the data for the selected batch or batches simultaneously differentiated. The data set for the multiple value curves may be compressed using a multi-variate algorithm into a small number of principal components and these may be shown on the parallel axis graph (PC1 to PC3).

Description

Multivarjate Visualjsatjon of a Batch Process This invention relates to

a method of visualising data for a batch manufacturing process.

In batch manufacture, many variables are often involved and their interaction can be complex. Conventional methods of analysing this data have limitations and the overall visualisation of the process is not possible. For instance, many values, such as product quality parameters, are recorded once per batch. Other values, such as a temperature trend, are recorded many times per batch. The batch duration may vary, precluding the possibility of simply comparing trends for a range of batches directly. Further, some values may interact with other values in a complex way. To overcome these problems, the present invention proposes a data analysis unit based on the parallel co- ordinate transform in which additional process values may be calculated using multi-variate algorithms and displayed in the same parallel co-ordinate graph. Hence, more data concerning batch production and quality variability can be displayed simultaneously.

The invention will now be described solely by example and with reference to the accompanying drawings in which: Figure 1 shows a parallel axis graph in which in-process measurements and final quality results are displayed, Figure 2 shows a time-based graph in which curves for a multi-value process parameter for multiple batches are displayed, Figure 3 shows two variables and their resultant principal component, Figure 4 shows the parallel axis graph with the addition of calculated principal components, Figure 5 shows the parallel axis graph as shown in figure 4 simultaneously with the time-based graph with selected data sets for both graphs shown as a thick line.

In figure 1, single values relating to batches are displayed as a parallel co-ordinate representation. In this example 10 variables are plotted, labelled Al to Al 0.These values may be in-process chemical or physical measurements (P1 to P5) or they may be final product quality measurements (Qi to Q5). Each line therefore represents one batch and the total display shows a representation derived from the data sets for multiple batches. For each variable a range of values may be specified and batches (B2, B4) having the specified variables (Q3 and Q4) within that range of values displayed with differentiation, enabling visualisation of the interaction of variables.

This visualisation has the limitation that each batch can be represented only by one value for each variable. In some cases a set of values for a variable may be needed to describe the batch history, for instance a temperature or pressure trend. In figure 2, the data describing a data trend for each batch is plotted against time.

Where the number of batches is large the trend graph can become difficult to interpret due to the number of data sets plotted on it and the fact that the overall time for each batch may not be identical.

The multiple-value trend data may be compressed using a multi-variate algorithm, for instance principal component analysis, into a small number of single values, which relate to the principal components of the data set. For instance, Figure 3 shows two process variables VI and V2, plotted at 900 to each other, with the values for multiple batches shown as a scatter graph. Each variable has a specification range, ViA to V1B and V2A to V2B. However, due to the influence of other contributions to overall variability, only those batches for which the values of VI and V2 are within the principal component area PC 1 will have optimal product quality.

For each batch, therefore, a number of multi-value variables (for instance temperature and pressure trend curves) can be compressed into a small number of principal components which are represented, for each batch, by a single value.

According to the present invention, the single values per batch holding the calculated principal components can now be plotted against the original single value variables in a parallel co-ordinate representation. Figure 4 shows the parallel axis graph as shown in figure 1, with the addition of further axes All to A 13 values comprising the calculated principal components PCi to PC3.

According to the present invention, multi-variate methods other than principal components but still compressing trend data into numbers which may be plotted in the parallel co-ordinate graph are included.

The display of the parallel representation holding the single-value per batch data including the calculated principal components and the time-based trend representation showing the multiple-value per batch data may be displayed simultaneously and selected batches shown with differentiation in both displays, as shown in figure 5.

Claims

Claims 1. Calculated principal component values which hold compressed

multiple-value data sets visualised with in-process and final quality single values using a parallel axis graph.

2. Multiple-value data sets plotted as a time-based graph together with simultaneous visualisation of single-value variables and principal components with data and data sets relating to selected variable values being shown with differentiation in both graphs.