Miha Sedej (2012) Steel sheet analysis with data mining models. EngD thesis.
There is more and more data being stored electronically nowadays to enable easy access, searching and processing of data. Various collections of data are being created for all aspects of our lives. These collections can provide us with new information if the modern techniques of data mining are applied. Companies can gain additional profit with these techniques, which is why this field of computer science is becoming more and more popular. In this thesis, collaboration was done with the Hidria Institute of materials and technology. The focus was on chemical and mechanical properties of steel sheets. The aim was to predict the hardness of these steel sheets from the other properties. Firstly, mechanical and chemical properties were determined, to conclude which contribute most to the prediction of hardness. Using different data mining methods, this data was then used as training samples for further predictions. Models consisting of all properties and only the best determined were compared. The Orange software package was used for data mining, which also provides a set of tools for data visualization. The performance of the method for automatic data visualizations search VizRank was tested in practice. Some of the most interesting visualizations found in data were shown. Methods for determining properties turned out to be useful as there was only a slight difference in accuracy between models built from all properties and only best scored ones. Prediction on the hardness was however, less successful. We detected a correlation between the chemical and mechanical properties and hardness, but accuracy was poor and thus not reliable enough for practical use. VizRank turned out to be very useful as it showed interesting correlations between data almost instantly, which would take a human many hours to find.
Actions (login required)