The Application Of SAE-Based Deep Learning Network In The Study Of Machine Tool Position Related Dynamic Characteristics

With the continuous advancement of technology,the manufacturing industry is gradually developing towards high speed,high precision and heavy load.As the core of manufacturing,CNC machine tools have further improved the performance requirements of CNC machine tools.The dynamic characteristics of machine tools have become the quality of high-precision machine tools.An important factor in the promotion.As a complex system composed of multiple parts,the machine tool has many influence factors on its vibration characteristics,and position and speed are the key factors affecting the machining accuracy.A large number of scholars have done considerable work in this respect,but they all pass different states of the machine tool.The modal parameters obtained by the traditional experimental mode or the running modal analysis are simply compared and summarized.The variation of the dynamic characteristics of the machine under the influence of different factors does not go deep into the mechanism level to discuss the reasons for the changes of the dynamic characteristics caused by different factors.This method not only requires a large number of experiments and experimental data analysis,but also takes time and effort,and the efficiency is low,and the analysis results are not universal,that is,the analysis results are only applicable to the test machine.Aiming at the problems existing in the above research,this paper introduces the deep learning algorithm to carry out experimental research on the dynamics of machine tool dynamics with position and velocity factors.This paper mainly does the following work in the following aspects:Based on the principle of random excitation,the random motion excitation sequences with different speeds were designed to run on the seven-axis five-link CNC machine tool.The modal information of the machine at different speeds was obtained.At the same time,the variable position experiment was carried out to study the modal parameters of different positions of the machine tool.The law of variation;according to the experimental results,the different influence mechanism of speed and position factors on the dynamic characteristics of machine tools is analyzed from the perspective of dynamics principle.The variable structure multi-degree-of-freedom simulation model is established,the influencing factors of the input excitation sequence of the simulation system are discussed,the excitation sequence parameters are determined,the modal mass distribution matrix is extracted as the position-sensitive eigenvalue,and the training data set is established using the two-layer SAE model combined with the BP neural network.The response is predicted to verify the feasibility of the method.The modal mass distribution matrix of the machine tool under dry running is obtained according to the mass matrix and the air-running mode.The three-layer SAE network is combined with the BP neural network to predict the impulse response.Then the LSCE algorithm is used to identify the modal parameters and the cosine is proposed.The comprehensive evaluation index is used to characterize the dynamic characteristics of the machine tool with position change.Finally,the effectiveness of the method is verified by DM4600.The research results show that the modal parameters of the machine tool are not affected by the running speed.The speed change mainly causes the change of the main vibration mode of the machine tool.The change of the machine tool position mainly causes the change of the kinetic parameter mass matrix "M" and the stiffness matrix "K".The simulation and experiments successfully verify the feasibility of the deep learning algorithm based on SAE and BP neural network in the study of machine position dynamic characteristics,and the rationality of cosine distance as a comprehensive evaluation index for the dynamic characteristics of machine tool with position change.At the same time,the multi-layer SAE model combined with the BP neural network deep learning method shows the ability to process large amounts of data efficiently and quickly in the research of machine dynamic characteristics,and provides an effective solution for online identification of machine tool dynamic characteristics.