PSI - Issue 28

Oleh Yasniy et al. / Procedia Structural Integrity 28 (2020) 1392–1398 Oleh Yasniy et al. / Structural Integrity Procedia 00 (2019) 000–000

1394

3

the process of plastic deformation of the matrix. The jump-like increments of strain occur when the stress exceeds the value  j . Therefore, the dependence of the strain on the stress is linear with a constant coefficient of proportionality Е  . As the stress increases by value of   , the next jump occurs. The whole tensile process within  j <  <  в is the alternation of parallel regions and momentary increments of strain, that is, the alternation of the processes of hardening and softening of the material. In particular, the relationship between the value of momentary increments of strain and the respective maximum tensile stress was found in Yasnii et al. (2002) (Fig. 1).

 , MPa

 р (  і )  р (  7 )  р (  6 )  р (  5 )  р (  4 )  р (  3 )  р (  2 )  р (  1 )

*

*

280

*

260

240

220

0,004

0,008

0,012

 (  i ), mm/mm

Fig. 1. Jump-like strain increments versus stress in AMg6 alloy specimens under tensile loading (data points correspond to the results of testing of 12 specimens under similar conditions).

Therefore, aluminum alloy AMg6 can be considered as a composite material consisting of a ductile matrix and brittle inclusions. The dispersoids in the AMg6 alloy act as the obstacles to the dislocation motion and contribute to the accumulation of the dislocations around them. When the dislocation density reaches a critical value, the brittle inclusions fracture, and thus the accumulated dislocation cloud disperses. The process of dispersoids’ fracture and dislocation cloud dispersion is accompanied by the respective plastic strain increment, that is, the deformation breakdown, which at the microscale is accompanied by the intensification of deformation in the planes of sliding and (or) the initiation of sliding on new systems. In the case of same size dispersoids fracture in the material, a momentarily increment of plastic strain occurs, which depends on the number of destroyed dispersoids n (  i ) and the mechanical characteristics of the material. 2.2. Methods of machine learning There are known the studies in which various methods of machine learning are applied to fracture mechanics tasks, in particular, to predict fatigue crack growth rate by Pidaparti et al. (1995), Mohanty et al. (2009), Yasniy et al. (2018). Therefore, the jump-like deformation of the AMg6 aluminum alloy should be predicted by the proposed algorithms of supervised learning, that is, by neural networks, boosted trees, support-vector machines, and k -nearest neighbors method. NN is one of the machine learning algorithms where the computer learns to solve different problems by analyzing examples of training dataset. In particular, they solve complex problems, such as classification, objects recognition, regression. Also, NN reveal efficiently the nonlinear relationships between variables. They can also predict new data that the system has not seen. That is, NN are the computational systems of interconnected neurons that functions similarly to the human brain. The multilayer perceptron is the simplest NN model, consisting of an input layer, one or more hidden layers of computational neurons, and an output layer. The first hidden layer receives a signal from the input layer, and, using synaptic weights, transmits the value forward to the next hidden layers, and up to the output of the network to obtain the result. In particular, to update the weights between each neuron, back propagation is used, which minimizes the prediction error.