Analysis On The Surface Defects Of Hot Strips Based On Data Mining Methods

More than one half of the steel products are plates and main of them are hot strips. But one of the biggest quality problems in hot production in Baosteel is surface defects. Statistical results showed that the reject caused by surface defects had already taken up 60% of total reject.As the surface defects can be caused by many problems in different locations and the locations may not always be the same, the defects will continue to happen if we can't locate the defects and solve the problem. So it's vital to analyze and locate the location and cause quickly to reduce the quantity of defects. At the same time, with the development of information technology the restriction of collecting, transferring, storing information and etc is loosed much more and Baosteel has accumulated much related data. But because of lacking proper methods and analysis tools, we have to find out the problem by manual analysis now. The effective use of information and efficiency of analysis have been the bottleneck to solve the problem.So this paper started the research work about surface defects. Its aim was to provide proper methods and tools for factories, improve surface quality management, find out the trend of surface defects in time, reduce the mass of defective products, perform quick analysis, find out the cause and accomplish the aim of discovering and taking measures early indeed.This paper introduced the characters of surface defects of hot strips and data mining thesis briefly at first. Based on this, we selected decision tree algorithm to analyze surface defects according to their multiform and complicated characters. So this paper analyzed some classic decision tree algorithm, compared their use when solving problems.Based on above we developed a decision tree system and improved some classic algorithms. During the software development we had used object-oriented technology and used standard template library and common tools to manipulate the data which had provided good interfaces for extending the model and facilitation for software maintenance. Compared with decision tree of business software, the system could be embedded into on-line system easily and used in much larger scope.At last we used the decision tree system to perform preparatory analysis on the line surface defect on the edge of hot strips and had given out the relationship between defects and technical factors in quantity which provided reference for further causal analysis and formed an elementary base for the research work of locating the defects.