2004 Volume 44 Issue 1 Pages 129-138
Development of flowery patterns or spangles on the surface of hot dip galvanised steel sheets is a common phenomenon. While elements like lead and antimony are known to be the primary factors contributing to spangle formation, sometimes they grow uncontrollably small or big. In this study, a data mining approach has been used to find a correlation between the spangle size in galvanised sheets, and the process parameters at one of the continuous galvanising lines at Tata Steel. All the process related data were collected from the CRM database, while the information on spangle size was generated through actual measurements. Statistical (factor analysis) and mining (neural classification mining) analyses were carried out. The most significant input variables with respect to spangle size were extracted. The artificial neural network classification model was developed using 849 records for training with a prediction accuracy of 57%. Strip thickness appears to be most sensitive on the spangle formation; whereas lead and antimony concentration in zinc bath, and the pressure difference between the top and bottom air knives seem to be more sensitive amongst the other eight significant parameters. The classification model can be used for prediction of spangle size given the process parameters. It can also be used as an important tool to set and adjust the process parameters to produce a given spangle size.