PSI - Issue 47

Zoltán Bézi et al. / Procedia Structural Integrity 47 (2023) 646–653 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

647

2

structures. Therefore, knowledge, accurate description and in-service monitoring and assessment of the dominant degradation processes in the structural materials of the equipment are essential for assessing the safety of operation and for evaluating the expected lifetime. The assessment of the integrity of nuclear power plant system components has traditionally been based on a fracture mechanics approach, according to existing codes and standards, which includes in-service inspections and estimation of radiation damage or examination of surveillance capsules placed in the reactor to predict the degradation. The degradation of the materials in the nuclear power plant was recognized in time, therefore monitoring programs were put in place to track the brittleness by means of Charpy test pieces placed in the reactor vessel. These surveillance specimens are running out, so it is necessary to test miniaturized specimens created from the used specimens. However, it is important to note that the standards and codes currently in use do not apply to the evaluation of miniaturized specimens. For this reason, Bay Zoltán Nonprofit Ltd. has set up a research project to develop a simulation procedure to evaluate cases not covered by the standard. In this paper, the determination of the material properties for a selected VVER440 power plant material and the optimization of the Gurson-Tvergaard-Needleman (GTN) damage parameters (Gurson (1977), Tvergaard and Needleman (1984)) using artificial neural networks (ANN) are presented. This method has already been applied by Aguir et al. (2010) and Shikalgar et. al. (2020) for other types of specimens however in case of flat small notched tensile specimens it has not been used so far. The determined damage parameters are applied to standard CT specimens, from which the fracture toughness (geometry independent) material property is estimated using finite element simulation software. During the determination of fracture toughness, the crack is propagated based on the virtual crack closure technique (VCCT) as it provides the possibility to calculate the J-integral in parallel with the crack propagation. 2. Determination of GTN parameters with artificial neural networks Neural networks consist of simple, parallel elements inspired by biological nervous systems. As in nature, the connections between elements have a strong influence on the functioning of the network. It is by adjusting the values of these connections (weights) that neural networks can be trained to perform a given task. Neural networks are trained so that the input has a specific target output. Figure 1 illustrates this setup, where the output and the target value are set until the output matches the target value. In general, this requires enough input values to determine the output as accurately as possible. Neural networks can also be used to solve complex problems that are difficult for humans to solve using traditional computer methods.

Fig. 1. Built-up of ANN.

In the following sections, the models that can be used for training and the function fitting will be presented, by which the GTN parameters are finally determined. 2.1. Training methods of ANN A very important element of the ANN method is the choice of the style of training, which can be divided into two main versions. For incremental methods, the weights and biases of the network are updated each time an input is given to the network, whereas for batch training, they are updated only after all the inputs have been imported.