Modelling Of Al-Si Cast Alloy Mechanical Properties With ANFIS

With high strength, low density, good castability and machineability, Al-Si series cast alloy has become one of the most important structural materials in manufacturing industry, and ZL101A is the most widely used. In order to improve the strength and ductility, it is important to develop model for predicting material properties and analysis the role of compositions and process parameters.Besides the ability of mapping the complex non-linear relationships between compositions and properties, the proposed adaptive neuro-fuzzy inference system modeling method could extract "IF-THEN" rules directly from data, and get knowledge hidden in the data. This overcomes the "Black Box" shortcomings of BP artificial neural networks, which widely used in modeling and predicting. Further more, ANFIS has higher modeling and generation ability than regress analyses and artificial neural networks.In this paper, based upon the accumulated product records and experimental data, the adaptive neuro-fuzzy inference system (ANFIS) method has been used to build mechanical properties models of Al-Si serises cast alloy. By the coupled use of adaptive fuzzy inference modeling and artificial neural network learning ability, a set of rules, which could help us better understanding the basic principles of alloying, have been generated directly from the experimental data. The developed fuzzy inference systems could mapping the relationships between compositions and mechanical properties accurately and could used to predict material propertieswith the known compositions and/or process parameters. With the help of the developed models, one could analysis the influences of compositions and/or process on properties, and get information for the further improving of material property. By integrating in the genetic optimization procedure, the fuzzy inference systems could be used to determine the optimal compositions for the highest or desired ultimate properties.The above ANFIS modeling method has successfully used in the property prediction of Al-Si series cast alloy. The methodology present here could also be used for other data based material property modeling and optimization practices.