Simulation Research Of Deep Learning In Wind Turbine Blade Structural Damage Identification

Wind turbine blade is one of the key components of wind turbine, which is prone to be damaged because of suffered alternating load and other kinds of factors. It has important meanings to ensure the normal operation of the wind turbine generator, and reduce economic losses, that wind turbine blade structure damage is detected, and the damage is located and quantified. Traditional nondestructive identification technology requires a priori knowledge, its identification precision is not high, and efficiency is low. Aiming at the shortcomings of the traditional nondestructive identification technology, a damage identification method is used widely that can establish the quantitative relationship between structural damage and its change of vibration performance by BP neural networks. Damage identification method based on vibration performance of wind turbine blade depends on accurate identification of the blade modal parameters. However, there is deviation between the measured value and the true value of the modal parameters, and emergence of false modal information due to environmental noise, test error and other reasons. This dissertation puts forward a thought to extract proper vector of wind turbine blade model parameters by deep belief nets(DBN), taking the proper vector of model parameters as input signal of BP neural networks, and blade damage state as output signal, to establish blade damage identification network, which can Reduce the influence of noise on the damage identification results, and improve the accuracy of the blade structure damage identification. The main research content is as follows:(1) The finite element model of wind turbine blade structure is established in ANSYS software, and multiple damage conditions that wind turbine blade may occur are simulated by reducing elastic modulus in proportion. The qualitative relationship between blade structural damage and modal parameters change is obtained, by modal analysis to compare the change of the natural frequencies and modal shapes of the wind turbine blade before and after blade damaged. The localization of the structural damage of wind turbine blade is realized by using the change rate of the element modal strain energy.(2) Network model of wind turbine blade structure damage identification based on DBN is established. The change rate of the element modal strain energy is used to be the characteristic quantity of damage diagnosis, to construct network training samples including single damage and double damage. The position and degree of wind turbine blade structure damage is detected effectively after study of the network. Meanwhile, the result got by above method is compared with what is got by the structure damage identification method based on BP neural networks in order to verify the feasibility of the method in this dissertation for improving the accuracy of fan blade structure damage identification.(3) After the wind turbine blade vibration identification experimental platform set up, artificial damage is simulated by opening a crack in the blade, and the blade vibration response data is collected. Then the auto-cross spectrum density method is used to identify model parameters of wind turbine blade. It is confirmed the method of this research is effective for damage identification of fan blade structure under laboratory conditions.