|
| 基于神经网络的金属塑性参数球形压入学习及反演识别 |
| SPHERICAL INDENTATION LEARNING AND INVERSION RECOGNITION OF METAL PLASTIC PARAMETERS BASED ON NEURAL NETWORK |
| 投稿时间：2024-08-12 修订日期：2024-12-03 |
| DOI： |
| 中文关键词: 球形压入，神经网络，金属，塑性参数，数值模拟 |
| 英文关键词:Spherical indentation, Neural network, Metals, Plastic parameters, Numerical simulation |
| 基金项目:国家自然科学基金项目(12272249, 12272256); 山西省基础研究计划项目(202203021211180, 202203021221159) |
|
| 摘要点击次数: 333 |
| 全文下载次数: 0 |
| 中文摘要: |
| 传统的力学测试方法需将试样加工成特定形状尺寸，不能满足特定应用工况下的测试需求，压入法则可实现原位测试且试样加工方便；而压入测试的局部作用方式使得其在获取材料力学参数时存在理论分析及表征上的复杂性，基于压入测试响应与神经网络算法可提供一种有效获取材料力学参数的新途径；本研究采用Instron万能材料试验机开展了紫铜、低碳钢的拉伸与球形压入测试，对所得压入载荷-深度曲线进行特征提取以作为后续数据基础；利用abaqus二次开发自动模拟不同塑性参数组合下的压入过程并获取压入载荷-深度曲线用于神经网络训练，对比不同寻找最优参数策略、激活函数、初始化神经网络参数方法并得到了具有良好学习效果且可用于有效反演金属塑性参数的神经网络结构；通过学习后的神经网络及实验所得压入载荷-深度曲线特征得到紫铜、低碳钢的相关塑性力学参数，将其与拉伸测试表征所得相应塑性参数予以对比，验证了本研究结合球形压入载荷-深度曲线与神经网络算法反演金属塑性参数方法的有效性，给出了一种有效获取金属/合金力学性能参数的新方法。 |
| 英文摘要: |
| Traditional mechanical testing methods require the specimen to be machined into specific shapes and sizes, which can not meet the test requirements of specific application conditions. Indentation method enables in-situ testing and convenient sample preparation, while the local loading mode of indentation leads to the complexity of theoretical analysis and characterization for materials mechanical parameters acquisition. Combing the indentation response and neural network algorithm can provide a new way to effectively obtain material mechanical parameters. In this study, tensile and spherical indentation tests of Cu and Fe have been carried out by the Instron universal material testing machine. The features of obtained indentation load-depth curves were extracted as the data basis for subsequent studies. Based on the secondary development of Abaqus software, a series of indentation numerical simulations with different combinations of plastic parameters were performed to obtain the corresponding indentation load-depth curves used for neural network training. By comparing the different optimal parameter finding strategies, activation functions, and methods of initializing neural network parameters, the neural network structure with good learning effect was determined to effectively obtain metal plastic parameters. Combining the indentation load-depth curve features form indentation test and the trained neural network, the related plastic mechanical parameters of Cu and Fe were obtained. Through the comparison of corresponding plastic parameters values from neural network learning and tensile test characterization, effectiveness of the proposed method for obtaining metal plastic mechanical parameters based on neural network algorithm and spherical indentation load-depth curve was verified. This study provides a new method for effectively acquiring mechanical properties parameters of metals/alloys. |
| 下载PDF阅读器 |
| 关闭 |
