Results prediction in industrial processes. A control tool based on the Knowledge

The difficulties involved in most metallurgical processes are well known, specially when the number of factors that act in them is very high. The problems are even more important if we want to forecast the process behaviour, because it is not easy to build a framework of links between the critical variables using the available information.

This work takes into account the availability of several computer generic tools which, with a suitable adaptation and endowed with the specific knowledge, are able of “learning” the process, of connecting a lot of facts and of forecasting the product quality, sustaining at the same time the process under control. These tools manage the plant information, help to reach a robust process, increase its knowledge and improve its performance, related with the reject level in ppm. The development of this type of tools, were considered some years ago as utopian.

The analytical method used is based in an initial selection of the defect that we want to study an the, factor or characteristics managing the process. Afterwards we will describe the most likely potential causes, source of the studied defect, and we will arrange them and give priority with probabilistic criteria, searching the root causes to all of them.

During the running of industrial processes we will connect, through the computer program, the experimental measurements of the selected factors with the actual results, so that the system learns, and at the same time we can reject the less significant variables, improving that way the reliability of the prediction.

The conclusions are based in real applications, put in practice in different lines of production, for the validation of the system and the testing of its efficiency using the corresponding success index.