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

1393

2

It is known that under the conditions of uniaxial tension test of a composite material, in particular, AMg6 aluminum alloy, the stress–strain diagram is observed at the soft type of loading by Strizhalo et al. (1997), (1993), (1999), Fedak (2003). The material failure that occurs in the structural elements has a “sawtooth” trend, as the fibers of the material are destroyed. The loading, which was perceived by the fiber of the material, is transmitted to the matrix. As a result, a tooth appears on the stress–strain diagram, which is proportional to the value of this load by Yasnii et al. (2004). The jump–like deformation studies of the aluminum alloy AMg6 during the tension test are usually quite expensive and complicated. Therefore, in the presence of experimental data, it is advisable to apply the methods of machine learning, in particular, neural networks (NN), boosted trees, support–vector machines (SVM), and k –nearest neighbors for data modeling by Hastie et al. (2009), Yasnii et al. (2018). The aim of this study is to compare the methods of machine learning and to select the final prediction model of aluminum alloy AMg6 stress–strain diagram. An important step of solving this task is to select the models that show the best results while predicting unknown data. The main criterion of the prediction quality is error. In particular, the prediction accuracy depends on the efficiency of the selected algorithm, how it is applied, the number of training data and the correlation between the actual and the predicted values.

Nomenclature  p (  i )

the stress

the stress of the start of the jumps process

 j

the stress increment

 

Е 

the coefficient of proportionality

  (  i ) the strain of the jump i the class of dispersions m the input i number j

 номер входу the number of the neurons in the layer

the number of the layer

l

m -th input signal of the j -th neuron in layer l synaptic weight of the m -th input j -th neuron of layer l

x mjl w mjl

signal NET of the j -th neuron of layer l

NET jl

OUT jl the output signal F

a nonlinear activation function threshold level of this neuron

 jl

2. Material and methods 2.1. Analysis of the stress-strain diagram of AMg6 material

Some structural materials, in particular, AMg6 alloy, are characterized by intermittent yield (jump-like deformation) under uniaxial tensile test conditions. The effect of low-temperature jump-like deformation of materials, which is associated with the initiation of a jump due to pulse influence, that is, thermal or mechanical, is important by Strizhalo et al. (1994). In particular, a well-known Portevin–Le Chatelie effect is the intermittent yield at relatively high temperatures, which is attributed to an abrupt increase in the number of mobile dislocations previously locked by impurities due to their release by Bernshtein et al. (1970). As a result of dynamic deformation aging, the process of dislocation locking by impurities and dislocation release recurs more than once, which is manifested by several “teeth” on the stress–strain curve. It is known that as the stress  p (  i ) increases, the strain of the jump increases. The region of the jump–like increments of strain under the soft type of loading is characterized by the stress of the start of the jumps process  j , the increment of the stress between the jumps   , the coefficient of proportionality in these regions Е  and the strain jump   (  i ) at the respective stress  p (  i ), where the symbol i defines the class of dispersions that are destroyed in