Revision 1 as of 2010-05-21 10:01:28
|Deletions are marked like this.||Additions are marked like this.|
|Line 35:||Line 35:|
|* [[IlmariJuutilainen|Ilmari Juutilainen]]|
VALTA (VALssauksen TArkentaminen)
01/2002 - 02/2004
VALTA is a large, co-operative with Rautaruukki Oyj, AvestaPolarit Stainless Oy and Imatra Steel Oy Ab, the largest steel manufacturers in Finland. The aim of the project is to improve the competitiveness of Finnish steel industry by increasing quality and delivery reliability of steel products. The aim is to be achieved by improving the control of the rolling process. Modelling and discovering the relationships between causes and consequences is the main contribution of the project. The project is funded also by TEKES.
The steel manufacturing process contains several phases from alloying to rolling and thus, the data collected for one steel product contains thousands of measurements made at different frequencies. All the process states have an impact on the quality properties of the steel product and usually not one but a combination of several factors will affect the quality. Discovering the useful information from the huge amount of data is not simple; knowledge about steel making, data mining, and statistical modeling tools is needed. The rolling process is fast which restricts the on-line -usability of the models.
Our research group attends on three sub-projects: The influence of heating on rolling, surface quality, and the homogeneity of mechanical properties. In all of these sub-projects we are in close co-operation with Rautaruukki. In the research we utilise our knowledge about data mining, knowledge discovery, soft computing, data pre-processing, and statistical models.
In the project we are researching also data pre-processing methods. The goal of the pre-processing is to reduce the data without losing any essential information. Statistical methods find the information easier from appropriately pre-processed data. The methods utilised in the pre-processing have been linear correlation and other standard statistical methods, parallel coordinates, k-means clustering and self-organising maps. Rolling conditions leading into retention of a steel product has been identified from the pre-processed data.
In the subproject the influence of heating on rolling the steel slab temperature after the pre-rolling is predicted using adaptive neural networks. The predictions are based on the measurements made from the furnace. The predictions can be used for optimising the heating and the rolling. The model reads data automatically from the process database and is adaptively and continuously working.
In the subproject surface quality reasons affecting on the forming of scale into the surface of hot-rolled steel strip are examined and modelled. The series of measurements about the process variables are compared to the location of surface detects observed by a machine vision equipment. During the rolling process a thick and short slab is rolled to a hundreds of meters long strip. The process variables, for example temperatures, are measured with short intervals across the whole strip. The data pre-processing problem is to locate the measurements made from different phases of the process to the ready strip.
In the homogeneity of mechanical properties sub-project the tensile strength and the yield strength of hot-rolled low-alloyed steel plates is modelled using heteroscedastic linear models. Heteroscedastic linear models predict not only the mean of strength but also the variance of strength. The models are utilised in a product planning tool, which approximates the probability of strength disqualification using the predictions of the models.