田小静,龚欢,杜宇,等.基于全连接神经网络的压铸件质量预测算法[J].精密成形工程,2024,16(3):159-164.
TIAN Xiaojing,GONG Huan,DU Yu,et al.Die Casting Quality Prediction Algorithm Based on Fully Connected Neural Network[J].Journal of Netshape Forming Engineering,2024,16(3):159-164. |
| 基于全连接神经网络的压铸件质量预测算法 |
| Die Casting Quality Prediction Algorithm Based on Fully Connected Neural Network |
| 投稿时间：2024-01-09 |
| DOI：10.3969/j.issn.1674-6457.2024.03.017 |
| 中文关键词: 压铸 MRMR相关性分析 神经网络 数据分类预测 归一化 |
| 英文关键词: die casting MRMR correlation analysis neural network data classification and prediction normalization |
| 基金项目:国家重点研发计划重点专项(2022YFB3706802) |
|
| 摘要点击次数: 820 |
| 全文下载次数: 443 |
| 中文摘要: |
| 目的 针对复杂压铸制造过程中高精度监控和质量预测问题，构建全连接神经网络，以提高压铸件缺陷分类和预测的准确性及高效性。方法 提出了一种基于全连接神经网络的算法，用于压铸件的质量预测。以汽车发动机下缸体为研究对象，先通过压铸岛采集关键工艺数据，后通过异常值处理和数据归一化进行数据预处理，再采用最小冗余和最大相关性的启发式算法(MRMR)进行特征处理，选出对压铸件质量影响较大的5个参数，该算法以3个压射速度、真空度、动模流量为输入层参数，以铸件质量为输出层参数。最后确定该算法的结构及各个参数，进行模型的训练与构建，并与不同算法进行性能比较。结果 与传统的决策树、SVM算法相比，该算法在相同数据集的分类和预测性能方面均更优，表明全连接神经网络在预测压铸缺陷方面具有优势。结论 该算法在实际应用中具有很大的潜力，证明全连接神经网络在预测能力和精度方面具有优势，可以为数据分类和预测提供更好的解决方案。 |
| 英文摘要: |
| The work aims to construct a fully connected neutral network to solve the problem of high-precision monitoring and quality prediction in complex die casting manufacturing process, thereby improving the accuracy and efficiency of defect classification and prediction of die castings. An algorithm based on fully connected neural network was proposed for quality prediction of die castings. With the automotive engine lower block as the research object, firstly, the key process data were collected through the die casting island. Next, the data preprocessing was carried out through the outlier processing and data normalization, and then the heuristic algorithm with minimum redundancy and maximum relevance (MRMR) was used for feature processing to select the five parameters with the greatest impact on the quality of die casting. The algorithm adopted the die casting speed, the vacuum degree and the moving mould flow as input layer parameters and the casting quality as output layer parameters. Finally, the structure of the algorithm and each parameter were determined, the model was trained and constructed, and the performance was compared with those of different algorithms. The classification and prediction performance of the algorithm on the same data set was better than that of the traditional decision tree and SVM algorithms, indicating that the fully connected neural network had advantages in dealing with the prediction of die casting defects. Therefore, the algorithm has great potential in practical applications, proving the advantages of fully connected neural network in prediction capability and accuracy, and can provide better solutions for data classification and prediction. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
渝公网安备 50010702501719号