Application Of Partial Least Squares And Artificial Neural Network In Predicting Of Hardenability Of Boron Steels

The hardenability is an important technological property of boron steels. It is connected with production and application of boron steels and development of new steel grade. Therefore, it is necessary to study the prediction of hardenability of boron steels, further, to promote the application and research of boron steels.PLS model and ANN model to study the relationship between hardenability and chemical composition of boron steels are established by PLS and ANN. At the same time, the establishment of PLS model and ANN model and theirs prediction results are analyzed.Hardness value of one point, hardness values of several points and hardenability curves of boron steels are predicted by ANN model based on the improved BP (Background Propagation) network arithmetic and it is much more accurate than ANN model based on the normal BP network arithmetic, experience formula and PLS model using the same set of data. Furthermore, the performance of ANN model based on the improved BP network arithmetic is examined with hardenability curves of 20MnVB steels. There is a good conformity of the prediction results with experimental results. The hit rate of boron steels' hardenability gets 90% when error of predicting values range is from -2HRC to 2HRC.The ANN model of prediction of boron steels' hardenability is used to analyze quantitatively the influence of alloying elements, e.g. Carbon, Silicon, Manganese, Vanadium, and Boron on hardenability of 20MnVB steel. In the range of chemical composition of 20MnVB steel given by technique standard (GB3 077-82), increasement of concentration of Mn can enhance steel's hardenability significantly while increasment of concentration of B and V has little effect on the steel's hardenability. The ANN model can also design the chemical composition of boron steels, which provides a new approach to develop the new steel grade.The forementioned results indicate that the accuracy of ANN model of prediction of boron steels' hardenability based on the improved BP network arithmetic is very good and it can be applied in practical production.In order to use the established ANN model conveniently, the hardenability prediction software of boron steels is programmed by means of MATLAB6.5.