By establishing the BP neural network model, the physical relationship between the structural parameters of the system and the dynamic parameters of the system (linear or nonlinear) can be reflected as the mathematical relationship between the network input and the network output of the neural network model, and the trained model is used for structural design. The re-modification, sensitivity analysis, and optimization are much faster than the optimization calculations based on other models (such as finite element models). However, how this idea is used for the design and development of machine tools: 2 (1) i1-2 (l fund project: Jiangsu Province 'Ninth Five-Year Plan) major industrial research funding (BG98006-2).
The rapid response system is still a problem worth studying.
Putting the BP neural network model together with the computer aided CAX (CAD/CAE/CAPP/CAM) information integration system into the database information management system developed by various manufacturing enterprises should be a better choice.
On the one hand, it makes up for the shortcomings of the traditional database information management system that users are difficult to intervene in the whole process.
Because in the use of CAD and CAE database, users lack the product modeling technology, finite element analysis technology or various comprehensive professional knowledge, so they can only passively use the database information management system to complete some simple query work. Modifications and analysis cannot be made for defects in the product. On the other hand, the user can directly call the data files of the trained neural network model in the database for rapid response design, so that the user can easily modify, sensitivity and optimize the parts even if they do not use CAD or CAE software.
1 BP neural network model of the main parts of the internal grinding machine The internal grinding machine is mainly composed of seven parts, such as bed, workbench, bridge board, box slide, bed head box, grinding frame skateboard and base. Firstly, the BP neural network model of each part is established separately, and the data files of the number of neurons, the weight coefficient and the threshold of each layer of the corresponding BP neural network model are obtained, and then placed in the corresponding address of the database.
In the modeling of the BP neural network model for the internal grinder parts, in addition to the method of selecting the training samples by the multi-level orthogonal table proposed in the paper, considering the internal grinding machine parts are three-dimensional entities, the sampling calculation of each sample It takes a lot of time. Therefore, instead of using the conventional method of performing 3D CAD modeling on CAD software and then calculating on CAE software, the modeling and calculation are all done on CAE software. This method can greatly reduce manual intervention and improve calculation. effectiveness. The calculation of all BP neural network samples in this paper is done under the large-scale finite element analysis software ANSYS. The modeling and calculations are all based on the APDL language of AKSYS. It can be completed by modifying the relevant structural parameters in the macro file each time.
After the BP neural network model of the machine tool parts is established, the user can modify the structure according to some simple operation instructions, and quickly obtain the dynamic characteristics of the parts after the structural parameters are modified. If the result is unsatisfactory, it can be modified repeatedly. Applying this method achieves a true rapid response design.
The block diagram of the database management system for the re-modification calculation of the internal cylindrical grinder based on the BP neural network model is given. Based on this, the sensitivity analysis and optimization calculation of the structural parameters can be further carried out on the parts.
2BP neural network model The application of the rapid response design of the internal cylindrical grinding machine is the structural diagram of the bridge plate, which is the layout diagram of the bridge ribs. Considering the symmetry of the bridge structure, the geometrical dimensions a and b of the rib layout can be determined as the design variables. The previous 4th modal frequency /, (i = 1, 2, 3, 4) is the output. Variable, establish 0 If the result is not satisfactory, modify the a, b value, re-run obviously, according to the sensitivity calculation formula, it is easy to find the sensitivity of the first, natural frequency to the roller / structure parameter (input parameter) S (j ), that is, (a), (b) gives the first 4 natural frequencies of the bridge plate. If the input parameters require the highest order of the first-order torsion frequency of the bridge, the finite element method is difficult to achieve. Even through the comparison of multiple programs, often only a relatively satisfactory solution can be obtained. Optimization on the neural network model can get the best results, and the computational workload is greatly reduced.
The model is calculated by using MATLAB's optimization toolbox, and the optimized calculation results are shown in Table 1. It can be seen from the table that, through optimization calculation, the first-order torsional frequency of the bridge plate is increased by 1289% compared with the initial value of the design variable, and the position of the rib plate obtained by optimization is not realized by the conventional scheme comparison optimization method. of. The optimization results of the design variables a, b are re-entered into the finite element model, and the error between the first-order frequency value and the BP model optimization value is less than 2%, which indicates that the BP neural network model of the bridge is truly reflected. The physical relationship between structural parameters and dynamic characteristic parameters is completely feasible for structural modification, optimization and sensitivity calculation on this model.
Table 1 Based on BP neural network model, the optimization results of bridge ribs position design variables / mma network output frequency / Hz finite element method calculation / / Hz error / initial value optimization results 3 Conclusion This paper introduces BP neural network model into database information management In the system, the user can easily modify, sensitivity and optimize the parts, which improves the problem of user participation and user requirements that are common in the current database information management system to a certain extent, and also participates in the parts of the user. Redesign creates conditions. The method of this paper has been successfully applied in the development of Jiangsu's "Ninth Five-Year Plan" major industrial research project / the development of a new generation of high-precision CNC high-precision inner circle.
(Finish)
Solar Panel Mounting Rack,Network Rack Profile,Nine Fold Rack,Sixteen Fold Rack
Unovo Machinery Co., Ltd. , https://www.rollformingprofile.com