Development Of The Grate-Kiln Pellet Expert System Based On VPRS

The grate-kiln method is a new kind of pellet production process regarded by iron and steel corporations in the world as soon as it comes out. Modern pellet production process is very complicated. Firstly, there are thermal radiation, heat conduction, and thermal convection in the system. Secondly, the process parameters and the operating parameters are strongly coupled with each other. Thirdly, there are many uncertain factors, such as pressure and fluctuations of heat value. So, the grate-kiln way of pellet production is a multivariable and strongly coupled process with time delays. Now, in the most factories, the workers control the system resorting to experience in manual state with great uncertainty. Once the thickness of the material or the temperature of the grate-kiln pellet change, and the workers can not adjust them in time, the quality of products and the reliable operation of equipment will be affected.For the above, the grate-kiln pellet expert system based on VPRS (Variable Precision Rough Set) is designed and developed to guide the pellet process in the thesis, aiming to reduce jamming in the process control and improve the quality of the finished pellets. The process of the grate-kiln pellet involve many parameters and huge volum of data, and the data mining method based on VPRS here is used to acquire knowledge from lots of data collected in the process and find the relationship between on-line measurement parameters and product quality. More over, the method can solve the bottleneck problem of the expert system to a certain extent.In this paper, the structure is as follows. First, the grate-kiln pellet method and the basic knowledge of expert system are introduced. Second, rough set theory is carefully discussed in the thesis, giving more emphasis on VPRS. VPRS model, introduces from the standard rough set model for the correct classification rateβ. Also, VPRS relaxes the demands of the approximate border of standard rough set model, strengthens anti-jamming capabilities of the rough set model and the predictive capability of new data. Therefore, this paper presents an algorithm for attribute reduction based onβsimilar set, core of the attribute conditions and attribute importantce of mutual information, which can be achieved by MATLAB programming. Based on this, the data mining based on VPRS is used to analyze the data in order to find the key factors which determine the finished pellets quality. As a result of the steps above, lots of production rules connecting with the process control are obtained. Third, with the above rules, the grate-kiln expert system is designed and built using Microsoft Access database and Microsoft Visual Basic6.0. There are three main functions in the system: quality prediction of finished pellets, control decision and quality evaluation. Finally, the expert system is tested. The results show that the grate-kiln pellet system is effective and has great value as a reference to the pellet production process control.