Calibration of Damage Parameters for Y5 and G91 Steels in Use of Genetic Algorithm

Determination of fracture toughness is prerequisite to perform elastic-plastic fracture mechanics assessment, for instance, leak-before-break analyses of nuclear piping systems and integrity evaluation of low uppershelf reactor vessels. However, sometimes, there are lacks of fracture toughness data especially for old vintage nuclear power plants and it is not easy to prepare standard specimens from archival materials or installed components. In these cases, damage mechanics is applicable as one of alternative approaches because several efficient models have been suggested to simulate ductile fracture behaviour during the last couple of decades. In the present paper, a multi-island genetic algorithm is adopted into well-known Rousselier model to resolve complexity of previous calibration methods since reliability of damage parameters is significantly dependent on the calibration method trial and error method, neural network method and so on combined with notched bar tests or small punch (SP) tests and analyzer’s experiences. SP test data of typical nuclear materials such as a low alloy steel (Y5) and a high Cr steel (G91) are used to determine damage parameters and, then, resulting values are applied to predict fracture toughness of the material. Load-displacement curves and fracture resistance curves are compared with those obtained from experiments, which show effectiveness of the proposed method.