| 马健,张宁,宁卫明,徐云杰,屈东山,王方俏,邬明宇,郑涛.基于深度学习的轮胎花纹预测噪声的方法研究[J].轮胎工业,2026,46(2):0112-0118 |
二维码（扫一下试试看！） |
| 基于深度学习的轮胎花纹预测噪声的方法研究 |
| Research on Methods of Predicting Noise with Tire Tread Patterns Based on Deep Learning |
| 投稿时间：2024-12-28 修订日期：2024-12-28 |
| DOI： |
| 中文关键词: 轮胎花纹 噪声 预测 深度学习 模型 神经网络 |
| 英文关键词: tire tread pattern noise forecast deep learning model neural network |
| 基金项目: |
| 中图分类号: |
|
| 摘要点击次数: 229 |
| 全文下载次数: 87 |
| 中文摘要: |
| 采用深度学习方法，通过训练轮胎花纹图像，建立轮胎花纹与轮胎噪声（简称胎噪）之间的非线性映射关系，实现对胎噪的准确预测，克服了传统胎噪预测方法依赖于经验公式和物理模型，难以涵盖轮胎花纹的复杂性和多样性的缺点。利用ResNet50，ResNet152，Inception V3，EfficientNet，MobileNet V3，Vision Transformer和ConvNext深度学习模型进行轮胎花纹预测胎噪试验，并对比各模型的预测性能。结果表明：ResNet152模型在≤0.5 dB和≤1.0 dB的误差区间的准确率分别达到72.22%和94.44%，但训练时间普遍较长；EfficientNet和MobileNet V3模型在小误差区间（≤0.5 dB）准确率较低，但参数量小且训练时间短，适合资源受限的高效应用；ResNet50和ResNet152模型在准确性和稳定性方面具有明显优势，其他模型在特定场景下可能需要进一步优化。 |
| 英文摘要: |
| By using deep learning methods to train tire tread pattern images and establish a nonlinear mapping relationship between tire tread patterns and tire noise，accurate prediction of tire noise was achieved，and the shortcomings of traditional tire noise prediction methods that relied on empirical formulas and physical models which were difficult to cover the complexity and diversity of tire tread patterns were overcomed.The tire pattern predicting tire noise experiments were conducted with various deep learning models such as ResNet50，ResNet152，Inception V3，EfficientNet，MobileNet V3，Vision Transformer and ConvNext，and the predictive performance of each model was compared.The results showed that the accuracy of the ResNet152 model reached 72.22% and 94.44% in the error intervals of ≤ 0.5 dB and ≤1.0 dB，respectively，but the training time was generally longer.EfficientNet and MobileNet V3 models had lower accuracy in small error intervals （≤0.5 dB），but with a small number of parameters and short training time，they were suitable for resource constrained and efficient applications.The ResNet50 and ResNet152 models had significant advantages in accuracy and stability，while other models might need further optimization in specific scenarios. |
|
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
