针对不锈钢焊缝缺陷特征提取存在主观单一性和客观不充分性等问题,提出一种融合迁移学习的AlexNet卷积神经网络模型,用于不锈钢焊缝缺陷的自动分类。首先,由于不锈钢焊缝缺陷数据较为缺乏,通过采用迁移学习对网络前3层冻结,减少网络对...针对不锈钢焊缝缺陷特征提取存在主观单一性和客观不充分性等问题,提出一种融合迁移学习的AlexNet卷积神经网络模型,用于不锈钢焊缝缺陷的自动分类。首先,由于不锈钢焊缝缺陷数据较为缺乏,通过采用迁移学习对网络前3层冻结,减少网络对输入数据量的要求;对后2层卷积层提取的特征信息批量归一化(batch normalization,BN),以加快网络的收敛速度;并使用带泄露线性整流(leaky rectified linear unit,LeakyReLU)函数对抑制神经元进行激活,从而提高模型的鲁棒性和特征提取能力。结果表明,该模型最终达到了95.12%的准确率,相比原结构识别精度提高了9.8%。验证了改进后方法能够对裂纹、气孔、夹渣、未熔合和未焊透5类不锈钢焊缝缺陷实现高精度分类。相比现有方法,其识别面更广,精度更高,具有一定的工程实践意义。展开更多
Large quantities of data are accumulated in process planning for body in white （BIW）. To acquire thepotential and valuable process knowledge from these data, the rough set theory and association rule technique arein...Large quantities of data are accumulated in process planning for body in white （BIW）. To acquire thepotential and valuable process knowledge from these data, the rough set theory and association rule technique areintegrated to discover the useful correlations between the welding type and process requirements. The correlationscan guide us to select the welding type according to the given process requirements. During data mining, everyprocess requirement is regarded as an attribute. First, the decision table for the welding type is constructed. Sec-ond, rough set theory is employed to remove redundant attributes. A simplified decision table is constructed.Third, association rule is used to extract the useful rules. Finally, an illustrative example indicates this methodol-ogy can extract useful rules for the selection of welding type.展开更多
文摘针对不锈钢焊缝缺陷特征提取存在主观单一性和客观不充分性等问题,提出一种融合迁移学习的AlexNet卷积神经网络模型,用于不锈钢焊缝缺陷的自动分类。首先,由于不锈钢焊缝缺陷数据较为缺乏,通过采用迁移学习对网络前3层冻结,减少网络对输入数据量的要求;对后2层卷积层提取的特征信息批量归一化(batch normalization,BN),以加快网络的收敛速度;并使用带泄露线性整流(leaky rectified linear unit,LeakyReLU)函数对抑制神经元进行激活,从而提高模型的鲁棒性和特征提取能力。结果表明,该模型最终达到了95.12%的准确率,相比原结构识别精度提高了9.8%。验证了改进后方法能够对裂纹、气孔、夹渣、未熔合和未焊透5类不锈钢焊缝缺陷实现高精度分类。相比现有方法,其识别面更广,精度更高,具有一定的工程实践意义。
基金support by Xinjiang special major project of science and technology [201130110]Xinjiang University doctor initial foundation [BS130119]
文摘Large quantities of data are accumulated in process planning for body in white （BIW）. To acquire thepotential and valuable process knowledge from these data, the rough set theory and association rule technique areintegrated to discover the useful correlations between the welding type and process requirements. The correlationscan guide us to select the welding type according to the given process requirements. During data mining, everyprocess requirement is regarded as an attribute. First, the decision table for the welding type is constructed. Sec-ond, rough set theory is employed to remove redundant attributes. A simplified decision table is constructed.Third, association rule is used to extract the useful rules. Finally, an illustrative example indicates this methodol-ogy can extract useful rules for the selection of welding type.
