| |
DataTools comprises a total of seven individual modules which build upon a project database (a type of DataMart) and exchange information among one another. Stored within the project database are not only the raw data requiring analysis, but also all the pre-processing steps and the interim results of the analysis. In individual cases it may be necessary to link the project database via a suitable interface to one (or several) central databases. This results in the following system structure: |
|
 |
|
|
|
The difference between DataTools and other data mining instruments is its particular suitability for evaluating data from technical processes and, especially, the possibility to analyse the influences which process and system variables have on product quality. With the aid of DataFeature, for example, it is possible to compile the in- and output variables from several, individual, serially connected processes in relation to the length of the end product and to save them in one single table. All of the data can then be suitably pre-processed (DataPrep) for subsequent, actual analysis.
The user interface is similar for all the modules, an example of which is given below (Figure 1):
|
|
|
|
Seen here is the result of a simple, linear calculation of the correlation between several influencing variables of the underlying process and a quality parameter (QP) of the end product. The linear correlation coefficients are rendered by means of a colour coding.
Also visible in this "scatter matrix plot“ are the non-linear interdependencies of the influencing variables and their relation to the quality information. As the quality information is made available here as discrete values (e.g. "good“, "moderate“, "poor“), it is necessary to select a different form for the appropriate diagrams (here: "supervision histogram“).
|
|
|
Displaying the product quality over the product length is a very important source of information. Shown in the adjacent diagrams is the result of an automatic surface inspection of cold-rolled steel and its aggregation over the strip width. Visible in the second diagram, besides the aggregated quality information, are several length-related signals from the originator process (in this case continuous casting). It is possible in this way to visualise several influences directly. DataView permits the display of one-, two- or three-dimensional quality information acquired as a function of length or time. |
 |
|
Besides graphical and simple statistical evaluation methods, the DataSelect, DataMod and DataClass modules of DataTools also contain more complex, multivariate and non-linear evaluation capabilities, the details of which are too extensive to be given here. Only the "decision tree" method will be mentioned as an example in this regard, which is capable of "translating" the relations between the data into verbal rules. The adjacent illustration shows examples of such "rules“.
DataTools can, if necessary, create models with on-line capability (e.g. for predicting a certain product quality) that can be integrated into any desired application.
|
|
|
|
The system is already in operation at several German and foreign steel enterprises and is marketed by the BFI in cooperation with a commercial software house and other partners. Offered besides the supply of the software are extensive training and support in the handling of projects.
Datasheet
| |
| |
