Available online at www.sciencedirect.com .-" -.;' ScienceDirect JOURNAL OF IRON AND SfEEL RESEARCH, INTERNPillONAL. 2009, 16(2): 80-83 Artificial Neural Network Modeling of Microstructure During C-Mn and HSLA Plate Rolling TAN Wen, LIU Zhen-yu , WU Di, WANG Guo-dong (The State Key Laboratory of Rolling and Automation, Northeastern University, Shenyang 110004, Liaoning, China) Abstract: An artificial neural network (ANN) model for predicting transformed microstructure in conventional roll- ing process and thermomechanical controlled process (TMCP) is proposed. The model uses austenite grain and retained strain, which can be calculated by using microstructure evolution models, together with a measured cooling rate and chemical compositions as inputs and the ferrite grain size and ferrite fraction as outputs. The predicted re- sults show that the model can predict the transformed microstructure which is in good agreement with the measured one, and it is better than the empirical equations. Also, the effect of the alloying elements on transformed products has been analyzed by using the modeL The tendency is the same as that in the reported articles. The model can be used further for the optimization of processing parameters, microstructure and properties in TMCP. Key words: artificial neural network; TMCP; microstructure; ferrite grain size Phase transformation is a very important process, and its prediction accuracy for ferrite grain size and transformed microstructure directly influ- ences the prediction accuracy of the mechanical prop- erties. There are always two ways to predict the transformation microstructure: (1) classical nuclea- tion and growth theory, (2) development of empiri- cal models. Both the methods have found industrial applications in the last decades-">". Many key pa- rameters need to be obtained through a large number of laboratory experiments when using the first meth- od. It is relatively easier to develop empirical equa- tions to describe the relationship between ferrite grain size and its inflaencing elements. Although many empirical equations have been developed to predict the ferrite grain size of C-Mn and high strength low alloy steel (HSLA) under conventional rolling and TMCP[I-7], their forms are various and are limited to a specific range, either C-Mn or HS- LA, and it is not easy to choose an appropriate one. Also, there is lack of empirical equations to predict the accurately transformed phase fraction. There- fore, it is necessary to develop the efficiency as well as a simple model that can predict both the ferrite grain size and the transformed microstructure of C- Mn and HSLA,. There is a strong effect of composition, cooling rate, and retained strain on the ferrite grain size. Composition is seen to affect both the base ferrite grain size and the contribution from the cooling rate, and any retained strain present in the austenite at transformation will refine the ferrite grain size. It has also been observed that the effectiveness of cool- ing rate in reducing the ferrite grain size decreases as the retained strain increasesl'", Therefore, because of its ability to process nonlinear multiple variables, an ANN model needs to be developed for the accu- rate mapping of the transformed microstructure and its influencing factors. Though an ANN model has been developed to predict the microstructure, it uses the measured austenite grain size as the input, and this needs a number of experimentsl'", In this arti- cle, a calculated a usteni te grain size with a retained strain will be employed in the development of an ANN model. Finally, the effect of alloying elements on the transformed microstructure has been analyzed Foundation Item-Item Sponsored by National Natural Science Foundation of China (50474086); Program for New Century Excellent Talents in University (NECT-04-0278) Blography:TAN Wen(l979-), Male, Doctor; E-DlBiI, tanwen_2001@sina.com; Revised Date, December 9,2006