压敏胶剥离强度预测中的数据增强技术

Data Augmentation Techniques in the Prediction of Peel Strength of Pressure Sensitive Adhesives

摘要: 配方是决定紫外光固化压敏胶(UV-PSA)性能的关键所在，因研究配方的传统方法难以获得丰富数据，限制了计算机技术在这方面的应用。为此，提出自适应合成过采样算法解决该场景下的数据稀缺问题。首先，通过距离度量策略预处理原始数据，使其适用于回归任务；其次，结合近邻与远邻策略以及非线性插值技术，生成具有多样化和代表性的合成样本；最后，利用扩展后的样本建立泛化能力强的支持向量回归预测模型。实验结果表明，增强后的UV-PSA的数据集提升了包括支持向量回归在内的所有模型性能，验证了提出的数据增强技术在UV-PSA配方研究中的有效性。

Abstract: Formulation is key to determining the performance of UV-PSA, and the use of computer technology in this area is limited by the lack of rich data available for traditional methods of studying formulations. Therefore, an adaptive synthesis oversampling algorithm was proposed to solve the problem of data scarcity in this scenario. Firstly, the distance measurement strategy was used to preprocess the raw data to make it suitable for regression tasks. Secondly, the nearest and far neighbor strategies and nonlinear interpolation techniques were combined to generate diverse and representative synthetic samples. Finally, the extended sample was used to establish a support vector regression prediction model with strong generalization ability. Experimental results show that the enhanced UV-PSA dataset improves the performance of all models, including support vector regression, and verifies the effectiveness of the proposed data augmentation technique in the study of UV-PSA formulations.