A deep learning model for online prediction of in-process dynamic characteristics of thin-walled complex blade machining

Online prediction of the dynamic characteristics of thin-walled workpieces, such as turbine blades, during the material removal process, plays an important role in the construction of digital twins systems for high-performance machining processes. However, the complex surfaces, thin-walled structures and time-varying characteristics of the blade machining process bring great challenges. The existing methods are either for simple structures or unadaptable to the continuous variation of the modal parameters, which cannot meet the requirements of online prediction for complex blade machining. To this end, this paper constructs a generative adversarial network with two output branches. By taking geometric information as input, online prediction of the modal parameters during the machining of complex thin-walled blades is realized. Considering the deviation between measured and predicted frequencies, an eXtreme Gradient Boosting model is established to modify the frequency branch of the network, which enables the model to be adaptive to machining uncertainties. By integrating the proposed network into the self-developed computer-aided manufacturing software, a digital shadow system of modal parameters prediction during blade machining is constructed. The verification experiments show that the calculation time of the proposed model is 1.35 s. The results demonstrate that the above system can achieve high-performance online prediction of modal parameters in the thin-walled complex blade machining process.