ISSUE 2/2014

/// 02
Saleem, M.; Malik, S.; Gottschling, J.; Hartmann, D.; Gemming, H. Intelligent process control in foundry manufacturing

In foundries, many parameters have an influence on the quality of the subsequent casting parts. With such a large number of influencing variables, it is extremely difficult to identify the cause-effect relationships clearly as well as rapidly and to develop the optimum process windows to avoid errors in most of the cases. In addition to the common complex intercorrelations between many processes and the quality-determining characteristics, there are also problems during collection of the required process parameters and the availability of appropriate data. For example: there are too few records in a data pool where the quality criteria are not met; the data are noisy; measured influence parameters are lacking; process specific values that are close to each other having strong deviations with respect to these quality criteria. In order to solve such problems, new software is developed within the scope of the industry alliance development proposal IPRO (FP7 EUROSTARS program, project E! 5092 IPRO; Intelligent Process Control in Foundry Manufacturing) based on machine learning methods that performs a comprehensive analysis and preparation of data before they are passed as input to the prediction tools. With such software, the quality of the casting products is to be predicted at a stage so that the process can still be influenced.
In this paper, the Analysis Software EIDOminer is described which supports an optimized parameter selection of a production process dynamically by using intelligent process data evaluation. The kernel of the software is the Intelligent Analysis Manager (IAM) through which, in case of a suitable data base, causal relationships between error patterns and the triggering process parameters can be identified which support to find an optimized process window. One of the advantages of IAM lies in the fact that it uses not only one analysis method of machine learning for the evaluation of the measured data, already edited in the pre-processor, but a variety of such methods which are summarized in a so called Functionbox. Thereby, the problem specific strengths of each analysis tool can be utilized and their weaknesses in the existing production process can be identified. The weaker tools, that is, which cannot predict good enough quality characteristics of the existing production process are switched off. The IAM communicates with the relational database EIDOfsdb (fsdb – foundry standard database), developed especially for the foundry processes, and the knowledge-based system EIDOwiba. The task of the EIDOwiba is to provide the most efficient parameter values, at each step, further to the production process. With this tool, a completely new quality of the process control in foundries and other manufacturing processes is possible.

››› Order magazine