Application of Neural Networks for Smart Tightening of Aeronautical Bolted Assemblies

Studies have shown that, at many points, the complexity of the behaviour of a bolted joint cannot be approximated without taking account of the interactions of the parameters of the bolt, the nut and the parts to be tightened. In addition, exogenous disturbances and the considerable number of combinations and configurations make it difficult to control the preload in bolted connections. As direct measurement of the preload of joints is impossible outside of laboratory tests, the aim of smart tightening is to use the torque and the tightening angle during the tightening phase to estimate the final value of the preload. The adaptation of parameters from an analytical relationship based on experimental tests is relevant, and the use of artificial intelligence with neural networks becomes interesting. This paper compares the performance of smart tightening based on a neural network with epistemic or arbitrary dispersions and conventional tightening. This strategy makes it possible to obtain an accurate estimation of the preload by taking epistemic dispersions into account during the training of the neural network.