Mechanical Stress Prediction of an Aircraft Torque Tube Based on the Neural Network Application

Abstract

The use of a predictive approach in the aviation industry is an important factor in aircraft maintenance. The main goal of this study was to create a new method for stress prediction during the operation of parts and to apply it on an aircraft torque tube (ATT). The method operates in real time during taxiing, takeoff, and landing using a neural network (NN). The stress calculated by the proposed method can be used in the future to calculate fatigue life and to save maintenance costs related to ATTs. This can play an important role in the evaluation of tests, such as unobserved crack failure. The main contribution of the presented methodology is in the fourth part of this study, where a new method of mechanical-stress prediction using a NN is described. The method essentially replaces finite element methods (FEMs), which require large amounts of time. The new method is much faster than commonly available methods, as the NN predicts the mechanical ATT stress in 0.00046 s, whereas the solution time using FEM is 1716 s for the same load step. In total, 36 regimes were calculated by FEMs in 17 h, 9 min and 36 s, whereas the novel method calculated the ATT stress for 36 regimes in 0.0166 s. The accuracy was also high, with R above 0.99. The main innovation presented in this study is the development of a method that can predict ATT stress in a very short time with a high percentage of accuracy and that can be used for stress and life prediction during the operation of parts. The partial results from the experimental tensile tests are also presented, and they are used for FEM calculations. The FEM results are used as inputs for the stress prediction by the NN.