Research On Strip Elongation Prediction Model In Annealing Furnace

Seam tracking of strips is the prerequisite for automation production of the continuous hot dip galvanizing line due to its precision contributing to the security of manufacturing equipment and the raw material consumption that directly determines the cost of productions. The strips would have length elongating and width narrowing attributing to the temperature and tension in annealing furnace used in the galvanizing line, which made it imprecisely for seam tracking. Therefore, the prediction model for strip elongation in annealing furnace, combing with the craft of the galvanizing line, was established. The research conducted by this thesis has significance in aspects of promotion of seam tracking precision, economization of raw material consumption, and reduction of the cost of production for the practical production realization in the future. The main works of the research conducting in this thesis include:(1) The data collected from the production line was used for analysis of the linear relationship between the elongation of strips and the factors contributing to it and a multivariable linear regression of them was established by general least squares (GLS) method. In terms of the collinearity between different factors that influenced, a regression model based on partial least square (PLS) was established. The simulation result showed the latter one based on PLS had both a higher prediction precision and reliability.(2) A model of radial basis function neural networks (RBFNN) with network center trained by a combination training method of subtractive clustering method and k-means clustering to resolve the problem of the k-means clustering method sensitive to initial value was applied. And genetic algorithm (GA) was applied to optimize the RBF center and width and a model of GA-RBFNN was built, which had center and width coded with real number separately that shortened the code length and promoted the calculation speed. The simulation result showed these two models had satisfied prediction precision providing the foundation for combination model establishment.(3) Aiming at the problem of narrow-range generalization and low reliability for the single model, a combination model of PLS regression, RBFNN trained by mixed method of subtractive clustering and k-means clustering and GA-RBF was established. The simulation result showed the combination model promoted not only the precision but also the reliability of the prediction.Compared the result with each model in analysis, combination model had the highest prediction precision. Utilizing combination modeling method to make use of the information varied in each individual model, the prediction precision of the model could be promoted effectively.