PSI - Issue 36

Iryna Didych et al. / Procedia Structural Integrity 36 (2022) 166–170 Iryna Didych et al. / Structural Integrity Procedia 00 (2021) 000 – 000

168

3

Fig. 1. Creep curve of AMg6 alloy: І – strengthening region, ІІ – steady creep region, ІІІ – jump- like creep region, ІV – failure region.

It is assumed that the magnitude of jump-like strain increment during creep  р is equal to jump -like increment of tensile strain  (  i ) . AMg6 aluminum alloy is characterized by intense hardening, therefore, in the first creep region the strain rate decreases quite intensely and reaches its minimum steady state. It should be noted that with increasing stress the time of transition to the region of steady creep decreases. The creep region consists of alternately variables regions of steady creep and regions of jump-like creep. Jump-like creep is characterized by a gradual increase of the creep rate, the following jump of deformation and the gradual leveling of the creep rate to a minimum level. The process of creep is completed by the region of neck formation and destruction of the material. In most of the short-term experiments of creep, the first three creep regions were implemented. In the process of creep there is one - two jump-like increment of strain (Fig. 2).

p, mm/mm 0,015

 с =  max = 340 М P а 1 -  а =  12,5М P а, f=50 Hz 2 -  а =  25М P а, f=25 Hz 3 -  а =  37,5М P а, f=16,5 Hz 4 -  а =0, f=0

1

2

0,010

3

4

0,005

100

200

300

400 t, min

Fig. 2. Diagram of jump-like creep (4) and dynamic creep (1-3) of AMg6 aluminum alloy.

The momentary strain increments have a larger scatter of values compared to uniaxial tension. This phenomenon is explained by the heterogeneity of AMg6 aluminum alloy by Yasnii et al. (2001).

2.2. Method of neural networks

Neural networks are widely used in studies of the materials strength and they are the modern method of machine learning, which solves tasks of mechanics with high accuracy by Kang et al. (2006), Yasnii et al. (2018, 2020). Attempts to reproduce the ability to learn and correct mistakes resulted in the creation of neural networks, which are the family of models built using the principle of organization and functioning of biological neural networks – networks of nerve cells of a living organism. The network is usually built using electronic components or simulated in software on the digital computer that performs the necessary calculations based on the learning process. To achieve good performance, neural networks use the interconnection of simple cells, i.e., neurons. One of the basic approaches in the area of data prediction is learning with a teacher. One interprets teacher’s participation as a target output that corresponds to certain input signals. Its aim is to minimize the signal error, which is the absolute value of the difference between the output signal and the target signal through the constant adaptation of neurons weights. In the course of learning, the data set is divided into two unequal parts, i.e., into training and test samples. Afterwards, using the training algorithm, the model parameters are set up using the training sample so that received as input data the model on an output would return data of the corresponding class. This approach is represented by a large number of models by Goodfellow et al. (2016), Haykin (2006), Smola et al. (2010). It is known that topology, algorithm of training and the functions of the neurons activation are the basic parameters of neural networks . In the current study, the sum of squares error function (SOS) was chosen and the