Cast quality and casting process cost control by data mining methods

Partners
Zakład Metalurgiczny Alstom Power - Elbląg
OŻ Drawski S.A. - Drawski Młyn
Silum Sp. Z o.o. - Opojowice
Specdodlew - Kraków, Radom
Huta Małapanew - Ozimek
OŻ Wulkan - Częstochowa

Aim of project
In the majority of foundry shops the data pertaining to casting process parameters are collected. Using this data for improving quality and reducing costs requires extracting from it information in the form of specific conclusions, rules or methods. This is made possible by the developments in the interdisciplinary area of study known as data mining, which employs statistical methods, artificial intelligence, data warehouse and visualization methods. The above methods come in many different kinds. However, their practical use is difficult due to the fact that their overall potential is not fully known. Moreover, the availability of tools facilitating easy industrial application seems quite limited. The aim of the project is to open the way to the use of data mining techniques in the casting industry. The following constitute the main aspects of the project: - an investigation and comparative analysis of the respective potentials of various data mining techniques, leading to the choice of the optimal methods, - the creation of the procedures of choice, initial analysis, and processing of production data, - the creation and testing of procedures and software designed for solving the most frequent kinds of problems encountered in casting. An important aim is the easiness of the actual application of the developed data mining techniques in foundry shops. Hence, the main emphasis of the project is on the assessment of the potential of already known methods, as well as on facilitating their actual industrial applicability. The idea is to employ the relatively simple software tools using the MS Excel platform – this should greatly facilitate the application of the entire system or its selected elements in the industrial practice.

Expected results
The project is aimed at creating the software applicable for the following tasks: - preparation of the data collected in the production process for the analysis, - visualization of the data and their correlations observable in the production process, - initial statistical analysis of the production data, - creation of nonlinear multivariate regression models with the use of selected self-learning systems, - extraction from models of the information pertaining to the relative importance of the input variables (process parameters), including whatever variable combinations (groups), with an indication of their potential interactions, - optimalization of process parameters based on regressive models, - time series data prediction, - creation of engineering knowledge in the form of logical rules, which will be applicable in designing and controlling production processes.