|
| 高强钢帽形梁零件冲压减薄预测分析 |
| Prediction and Analysis of Stamping Reduction of High Strength Steel Cap Beam |
| Received:October 07, 2021 |
| DOI：10.3969/j.issn.1674-6457.2022.04.007 |
| 中文关键词: 人工神经网络 贝叶斯优化 帽形梁 交叉验证 |
| 英文关键词: neural network Bayesian optimization cap beam cross validation |
| 基金项目:湖北省教育厅科学研究计划(B2020245);湖北省重点研发计划(2020BAB140);武汉市科技成果转化专项(2019030703011511);中央高校基本科研业务费专项(213107006) |
| Author Name | Affiliation | | PANG Qiu | School of Mechanical and Electrical Engineering, Wuhan Donghu University, Wuhan 430212, China | | LUO Bo-feng | Hubei Key Laboratory of Modern Automobile Parts Technology, Wuhan 430070, China
Hubei Collaborative Innovation Center for Auto Parts Technology, Wuhan University of Technology, Wuhan 430070, China | | WANG Jun-jie | Hubei Key Laboratory of Modern Automobile Parts Technology, Wuhan 430070, China
Hubei Collaborative Innovation Center for Auto Parts Technology, Wuhan University of Technology, Wuhan 430070, China |
|
| Hits: 2107 |
| Download times: 930 |
| 中文摘要: |
| 目的 针对高强钢帽形梁强度高、塑性差、结构复杂、冲压过程中出现减薄破裂等情况，建立QP980高强度钢冲压成形减薄预测模型，解决实际生产工程难题。方法 以典型高强钢车身零件帽形梁为对象，利用人工神经网络模型研究工件结构和减薄率之间的关系，将贝叶斯优化算法和循环人工神经网络相结合，建立冲压成形减薄的高精度预测模型，对高强钢帽形梁零件冲压成形时减薄量进行优化设计，通过AutoForm软件验证算法模型的准确性。结果 拉深高度对减薄率影响最大，对外减薄率影响达到41.7%，对内减薄影响达到46.2%，人工神经网络模型对测试集5组数据的预测平均误差均小于0.3%。根据人工神经网络求解QP980钢在极限减薄率25%下的最大拉深高度为55.417 mm，人工神经网络预测结果与Autoform仿真结果相差0.3%，验证了人工神经网络模型的准确性。结论 采用该模型解决了CAE模拟在较少试验数据条件下算法预测精度差的问题，能有效缩短高强钢零件冲压成形调试周期，提高生产效率。 |
| 英文摘要: |
| The work aims to establish a stamping reduction prediction model of QP980 high strength steel in view of the high strength, poor plasticity, complex structure, reduction and cracking defects in stamping process of high strength steel cap beam, so as to solve the difficulties in practical engineering. With the cap beam of typical high strength steel body parts as the object, the relationship between workpiece structure and reduction rate was studied by artificial neural network model. A high-precision prediction model of stamping reduction was established through combination of Bayesian algorithm and recurrent artificial neural network optimization hyperparameter, and the accuracy of the algorithm model was verified by AutoForm software. The tensile height had the greatest effect on the reduction rate, reaching 41.7% for the external reduction rate and 46.2% for the internal reduction rate. The average prediction error of the artificial neural network model for five groups of data in the test set was less than 0.3%. According to the artificial neural network, the maximum tensile height of QP980 steel under the limit reduction rate of 25% was 55.417 mm, and the difference between the prediction result of the artificial neural network and the Autoform simulation result was 0.3%, which verified the accuracy of the artificial neural network model. This model solves the problem of poor prediction accuracy under the condition of CAE simulation and less experimental data, which can effectively shorten the debugging period of high strength steel component stamping and improve production efficiency. |
|
View Full Text
View/Add Comment Download reader |
| Close |
|
|
|
渝公网安备 50010702501719号