Aerospace Technic and Technology

An approach based on using the neural networks has been developed for predicting the fatigue strength of gas turbine engine (GTE) parts at the manufacturing stage. The neural network model implemented in the Delphi programming environment consists of an input layer containing four elements, one hidden layer with four neurons, and an output layer with one element. The output parameter of the neural network is fatigue strength, and the input parameters are determined by the technological process of manufacturing parts. These are the surface roughness, the degree, and depth of strain hardening and the residual compressive stresses in the surface layer. The sigmoid function is used as the activation function of the neural network. Back propagation error algorithm is developed to network’s training. The neural network is trained according to a previously prepared training and test data, including experimental and literary data. The fatigue strength values of GTE parts were determined for a constant base and test conditions. The mean squared errors calculated by the neural network for the training and test samples are 0.021 and 0.034, respectively. The input parameter ranges, on which the neural network model correctly performs prediction of fatigue strength, were determined. The relationships between the significance of each of the network input parameters for assessing the complex effect of surface layer parameters on fatigue strength are established. It is shown that residual compressive stresses and the degree of surface hardening have the greatest influence on the increase in the endurance limit of parts after processing. The possibility is considered and examples of the solution of the inverse problem are given, when by known values of the fatigue strength the values of the input parameters of the neural network, leading to the appearance of the available output, are searched. The developed approach may allow even at the stages of technological preparation of production to provide the required value of fatigue strength, to minimize the amount of testing and make the choice of optimal processing modes for GTE parts for the most effective hardening.

fatigue strength; surface hardening; residual compressive stresses; strain hardening; surface roughness; neural network

Gorelov, V. A., Kudrov, A. S., Merkulova, N. S., Yakovlev, M. G. Issledovanie vliyaniya ostatochnykh napryazhenii na ustalostnuyu prochnost' obraztsov ikh titanovykh i nikelevykh splavov [Investigation of the effect of residual stresses on the fatigue strength of titanium and nickel alloys samples]. Problemy mashinostroeniya i nadezhnosti mashin, 2014, no. 5, pp. 47-51.

Sakhnyuk, N. V., Tsyganov, V. V. Vliyanie parametrov kachestva poverkhnostnogo sloya na vynoslivost' lopatok kompressorov GTD novogo pokoleniya [The influence of quality parameters of the surface layer on the endurance of a new generation GTE compressor blades]. Vіsnik SevNTU: zb. nauk. pr. [Bulletin of the SevSTU: col. scient. papers], 2011, vol. 117/2011, pp. 153-156.

Kachan, A. Ya., Ulanov, S. A. Matematicheskoe modelirovanie vliyaniya tekhnologicheskoi nasledstvennosti finishnykh metodov obrabotki na predel vynoslivosti detalei GTD [Mathematical modeling of the influence of technological heredity finishing methods of processing on the endurance limit of GTE parts]. Vestnik dvigatelestroeniya, 2015, no. 1, pp. 81-86.

Biblik, I. V. Prognozirovanie ustalostnoi prochnosti poverkhnostno uprochnennykh materialov na osnove neirosetevogo modelirovaniya [Prediction of fatigue strength of surface-hardened materials based on neural network modeling]. Vestnik dvigatelestroeniya, 2018, no. 2, pp. 82-86.

Petukhov, A. N. Soprotivlenie ustalosti detalei GTD [Fatigue Resistance of GTE Parts]. Moscow, Mashinostroenie Publ., 1993. 240 p.