9361

8643

كارشناسي ارشد

مواد- شناسايي و انتخاب مواد

اصفهان: دانشگاه صنعتي اصفهان، دانشكده مواد

1393

چهارده،109ص.:مصور،جدول،نمودار

ص.ع.به فارسي و انگليسي

احمد رضاييان، عباس نجفي زاده

قاسم ديني

18/9/93

محمدرضا طرقي نژاد، علي اشرفي

مهندسي مواد

ID8643

111 Application of artificial neural network modelling for predicting TWIP steel properties Reza Pahlavan Isfahan R Pahlavan@ma iut ac ir Date of submission Department of Materials Engineering Isfahan University of Technology Isfahan 84156 83111 IranDegree M Sc Language FarsiSupervisor A Rezaeian assistant Professor a rezaeian@cc iut ac irA Najafizadeh Professor a najafi@cc iut ac irAbstractIn the recent years the metallurgy of high manganese steels especially twinning inducedplasticity steels TWIP and Transformation induced plasticity TRIP has been consideredas an important scientific issue The excellent combinations of Due to the appropriatechemical structure these steels possess desirable strength and ductility High Mnaustenitic TWIP steels provide a great potential in industrial applications especiallyfor structural components in the automobile industry owing to their excellentcombination of strength and ductility It has been reported that such combination isattributed to austenite matrix as well as twinning process during plastic deformationIn recent years artificial neural network has been widely used to predict properties ofmaterials without using expensive and time consuming trial and methods Therefore inthis study artificial intelligence was employed to study and predict the mechanicalproperties of TWIP TRIP steels Input variables for the neural network were chemicalcomposition parameters weight percent of manganese aluminum and silicon andthermomechanical parameters such as annealing temperature annealing time and thepercentage of cold work the neural network was designed and trained in such a way thatcould predict the effect of above mentioned parameters on output and favored variablessuch as yield strength tensile strength and percentage elongation Feeding data for thenetwork was extracted from the credential sources and papers Among the data 20 wasassigned for test stage 20 for verification and the remaining 60 was used for trainingthe network Neural network program was developed separately for each categories ofchemical and mechanical parameters Three separate programs were designed to achievemore accurate results for each group of parameters Also two methods i e back propagation error and radial basis were utilized during the neural network By using theneural networks influence of each parameter on the mechanical properties wereinvestigated and the weught percent of each impact was determined The results indicatedthat the trained model could precisely predict the sensitivity of the mechanical propertiescorresponding to the input variables In the next step using the results of neural networks genetic algorithms equations for the model were estimated by which the results obtainedfrom the neural network were optimized Keywords high manganese steels TRIP TWIP mechanical properties chemical parameters thermomechanical parameters neural network genetic algorithm

احمد رضاييان، عباس نجفي زاده

قاسم ديني

محمدرضا طرقي نژاد، علي اشرفي