| Summary: | The paper considers the coefficient inverse problem of identifying the creep characteristics for metal structures deformed under high-temperature creep conditions. Creep deformation is widely used in various engineering problems, including the shaping of structural elements and the calculation of long-term strength. The creep phenomenon is exhibited by most known materials, including metals, concrete, and composites. Herein, there is no unified creep theory, which is explained by the complexity of the described process. The constitutive creep equations, as a rule, contain a set of tunable parameters (material constants) necessary for identification. There are many methods for determining material constants, but most of them are suitable only for specific creep models. In previous works, the authors proposed a method of multilayer functional systems, which in combination with neural network technique led to an original approach in the application of deep learning neural networks to the construction of approximate solutions of differential equations. Such approach allows combining the flexibility of artificial neural networks and the performance of traditional numerical methods for solving inverse and Cauchy problems. With regard to the creep models identification problem multilayer method allows naturally obtain a deep network, giving an approximate solution to the problem under consideration with an explicit occurrence of material constants. The identification of material constants is reduced to approximating experimental data, for example, by the least-squares method using the constructed multilayer solution. The advantages of the multilayer method are shown in solving the identification problem of the material constants for stress range. We selected the creep model describing the tension of cylindrical specimens of steel 45 at a constant stress level and temperature T = 850°С as a test problem for our multilayer method. Authors carry out analysis of obtained results and a comparison of the calculated data with the experiment and the results of other authors to confirm the reliability of the method used.
|
|---|
