CNC Machining Parts is the use of cnc machining out, including: drilling, boring, cutting, turning, screw. Product material are:aluminum(6061,6061-T6,6063,6083,7075.5052,6082,etc),stainless steel(303,304,316,316L,etc),carbon steel,brass,copper,titanium alloy,plastic .In the high-speed machining out of the workpiece has a good finish. cnc machining products surface treatment has anodized(black, red, blue, yellow), electroplating, sand blasting, laser engraving LOGO, powder coating,plating,brushing, rolling flower, according to the special requirements of the drawings.We have a special inspection department to carry out inspection, to ensure that each size of products meet the requirements of the drawings,cnc machining parts is the most common in mechanical processing, low density, not easy deformation, high precision products. So it is widely used in various industries. CNC Machining Parts CNC Machining Parts,CNC Machine Parts,CNC Machine Auto Car Parts,CNC Machining Agricultural Spare Parts YaoPeng Metal Products Co., Ltd. , http://www.ypmfg.com
Manufacturing Automation Mao Haijun Sun Qinghong! Chen Nan! , Chen Xin! He Jie! Wu Zhiyue, Zheng Wenyou, Wang Jianping! Department of Mechanical Engineering, Southeast University, Jiangsu Nanjing Wuxi Machine Tool Co., Ltd. Wuxi neural network based grinding machine spindle system dynamic optimization method introduction With the development of computer technology, finite element method has become the main means of large-scale complex structural dynamic analysis, but computer computing The contradiction between the ability and the complexity of the actual structure is still the key factor that constrains the problem solving. Especially when the dynamics of large structures are optimized, the work efficiency is low. Even if the sensitivity analysis method is used, it is still necessary to step through multiple design variables. Searching for excellence, there are a lot of double counting, it is difficult to achieve rapid response to the design. In recent years, the emerging neural network method uses its highly parallel information processing, strong nonlinear mapping and adaptive learning ability to provide a means to eliminate this long-standing constraint.
The basic idea of ​​the method is to linearly or nonlinearly reflect the physical relationship between the system structural parameters and the system dynamics parameters as the mathematical relationship between the network input and the network output of the neural network model. The process of establishing this relationship is the process of establishing the neural network model. Call it training. It is straightforward and simple to use the trained model to re-modify and optimize the structural design. The calculation speed is much faster than the optimization calculation based on other models such as the finite element model. The processing idea of ​​such an optimization problem is particularly adapted to the basic topology form of the structure has not changed much, and some basic size parameters often need to be optimized in the design and development of the rapid response system application, machine tool design is the field of such demand.
The key to neural network modeling is training, and the number of input and combination of samples increases sharply as the number of input parameters increases during training. This brings a lot of work to neural network modeling, and even cannot be achieved. To this end, this paper proposes a method for selecting neural network training samples using multi-level orthogonal tables. The trial results show that it is only necessary to select the fund project 'Jiangsu Province's 95 major industrial projects. )) / (0 Author brief introduction 'Mao Haijun ―, male, Yuyao, Zhejiang, associate professor, Ph.D. student, School of Transportation, Southeast University, research direction: mechanical structure optimization design and neural network training sample method, to ensure sufficient modeling accuracy conditions Under this method, the modeling workload of neural network is greatly reduced, so it has great practical value. Based on this method, the neural network model of the grinding machine spindle system is established and the structure modification and optimization calculation are carried out. The analysis and optimization of large complex structures has the advantages of simplicity and efficiency.
In the mechanical model of the manufacturing automation table grinding machine spindle system, the grinding head can train a higher precision neural network for a much smaller number of arranged samples of the pulley. Based on this method, the neural network is used to study the dynamic analysis of the structure. Firstly, the neural network model of the grinding machine spindle system is established, and then the structure modification and optimization design is carried out based on this model.
Neural Network Modeling of Grinding Machine Spindle System Establishing a neural network The selection of its sample is critical. It is related to whether the established network model can correctly reflect the relationship between the input and output variables selected by the actual physical system. Similar to the usual interpolation, the usual practice is to evenly distribute the input variables in the interval of interest. If there is an input variable, each variable takes a uniform value, that is, each variable has a horizontal number, according to the arrangement. There should be a training sample for the combination. This value is sometimes very large, and even a very simple structure is impossible to perform all calculations. To this end, the author uses orthogonal table to study this problem, and finds that for a single input variable, when the horizontal number takes a larger value, the network model obtained by training in the orthogonal table still achieves satisfactory accuracy. Time! . In this way, the neural network modeling of the structural system has practical feasibility. The following neural network modeling of the grinding machine spindle system applies a multi-level orthogonal table to select training samples.
The figure shows the mechanical model of a certain type of grinding machine spindle system in the feed plane. Take the angular displacement stiffness of the support)''. For the input variable, take the front-order frequency as the output variable to establish the neural network model in the input variable allowable region. The range of support stiffness coefficients is determined by the measured transfer function method.
The number of hidden layer neurons in neural network modeling can theoretically be determined according to the theorem. However, in practical applications, since the number of neurons in different hidden layers is selected to have different effects on the training error and test error of the network, the number of neurons in the hidden layer is often determined through multiple trial calculations. In the input variable and the previous-order frequency network model, the number of neurons in the input layer, the hidden layer and the output layer are respectively, and in the neural network model of the input variable and the point static deflection, the input layer and the hidden layer are The number of neurons in the output layer is respectively and '. The built neural network model has an input variable, and each variable is uniformly valued. That is, each variable has a horizontal number. , there should be a training sample from the permutation combination, it is not feasible to train such a number of samples. Now use the orthogonal table to take only one sample for training. Here, the expected output value of the sample is calculated by the finite element method, and the training uses the weight coefficient and threshold correction formula of the literature 23.
In order to test the generalization ability of the network model, multiple samples were selected in the area allowed by the input variables. The results show that the test output calculated by the neural network model is consistent with the expected output value, and the error A. The accuracy of the mold is within the limits allowed by the project. It is indicated that the neural network model constructed has well described the mapping relationship between structural parameters and structural characteristic parameters of the grinding machine spindle system. The modified reanalysis and optimization calculations of the model can be carried out in the established neural network model without having to use the original finite element model. Therefore, the subsequent optimization process will become straightforward and simple.
In the mechanical model of the grinding machine spindle system shown in the neural network model of the grinding machine spindle system, the dynamic model is modified by the network with the natural frequency of the order.
Since the natural frequency of the spindle system is a function of the stiffness of the support, the model can be modeled close to the real model by correcting the stiffness of the support. The support displacement stiffness design variable is selected, and the sum of the squared natural frequency calculated values ​​of the spindle system and the corresponding measured frequency of each step is the minimum. The selection of the sequence and the optimization method of the sub-planning are used to calculate the optimization calculation results. Automated by the watch manufacturing! References Xu Yanshen mechanical dynamic design Beijing 'Mechanical Industry Press (Jiao Licheng neural network system theory Xi'an Xi'an University of Electronic Science and Technology Press, Cong Shuang face / 0 toolbox neural network theory and application Hefei China University of Science and Technology Press practical guide Xi'an Northwestern Polytechnical University Press ()) 1 (4 table grinder spindle system neural network model support stiffness correction results table based on neural network model based grinding machine spindle system optimization results show that by modifying the stiffness of the support stiffness network output is almost the corresponding The measured frequency indicates that the neural network model at this time is very close to the real model of the structure.
Optimization of Grinding Machine Spindle System Based on Neural Network Because the modified neural network model of the grinding machine spindle system is very close to the real model of the structure, the model can replace the original finite element model. In this way, the optimal design of the spindle system can be performed on an established neural network model.
Since the neural network model actually reflects the physical relationship between the system structural parameters and the system dynamic characteristic parameters as the mathematical relationship between the network input and the network output of the neural network model, the optimization calculation on the neural network model is more limited than The metamodel is much easier, and this feature is particularly prominent in large and complex structures.
For the grinding machine spindle system shown in the figure, it is required to determine the length and diameter of the specified shaft segment when the spindle weight is the lightest under the condition that the first-order frequency of the spindle is not lower than the point static deflection. The optimization method of the sequence sub-planning is still used for calculation, and the table shows the optimization calculation results. It can be seen from the table that the optimization effect is very significant.
Conclusion This paper proposes a method for selecting neural network training samples using multi-level orthogonal tables. According to the method of training, as long as a small number of arranged samples are selected, a network model with satisfactory accuracy can be obtained, which makes the neural network modeling of the structural system practical.
Based on this method, a neural network model of the grinding machine spindle system is established. Because the built neural network model has high precision, it can well describe the mapping relationship between the structural parameters of the spindle system and the structural characteristic parameters. Therefore, the neural network model is modeled according to the measured frequency of the experiment. Correct and optimize calculations.
Because the neural network model linearly or nonlinearly reflects the physical relationship between the system structural parameters and the system dynamics parameters as the mathematical relationship between the network input and the network output of the neural network model, structural modification and optimization on the neural network model It is more straightforward, simpler and more efficient than on other models.
The method of this paper has achieved good results in the dynamic modeling of the connection between the spindle system of a certain type of grinding machine and the structural parts of the bed. It shows that the calculation of neural network is combined with the traditional numerical method, which is of great practical significance for the calculation of large complex structures.