基于自适应共振理论的结构损伤识别
|
摘要
目的基于自适应共振理论,提出一种基于ART2神经网络的结构损伤识别方法,以实现结构损伤识别的自主学习.方法采用一种改进算法来解决ART2方法中对输入矢量必须是非负实数的要求,并通过主成分分析方法对网络的输入矢量进行降维处理.结果通过对健康监测基准问题模型的计算表明,所采用的改进算法使得网络的输入扩展到整个实数域,且主成分分析方法有效地降低了输入矢量的维数,减少了网络的学习训练时间,从而提高了网络的泛化和判别决策能力.结论基于ART2神经网络的结构损伤识别方法具有自组织、反馈式增量学习机能,能够在不破坏原有记忆样本的情况下,学习新的样本,可以在较强噪声环境下快速准确地识别损伤,适宜于结构损伤的在线监测.
An ART2 neural network based on adaptive resonance theory is put forward in this work to identify the damage of the structures and to realize the on-line self-study of the network.An improved algorithm and the principal components analysis method are adopted into the network to improve its ability in damage identification of the structures.In order to validate the effectiveness of this method,the numerical sample about the benchmark problem is analyzed by using the network under different noisy conditions.The results show that the improved ART2 algorithm can expand the input vectors from the nonnegative real number field of the standard ART2 to the whole real number field,which can broaden the damage identification scope of ART2.The dimensions of the input vectors are reduced by the principal components analysis,which can effectively shorten the training time of the network.It can be concluded from the numerical analyses that the improved ART2 neural network can be used to study the new samples itself without destructing the original samples.It can recognize the damage patterns under strong noisy environments rapidly and precisely,and it is also applicable to on-line monitoring.
An ART2 neural network based on adaptive resonance theory is put forward in this work to identify the damage of the structures and to realize the on-line self-study of the network.An improved algorithm and the principal components analysis method are adopted into the network to improve its ability in damage identification of the structures.In order to validate the effectiveness of this method,the numerical sample about the benchmark problem is analyzed by using the network under different noisy conditions.The results show that the improved ART2 algorithm can expand the input vectors from the nonnegative real number field of the standard ART2 to the whole real number field,which can broaden the damage identification scope of ART2.The dimensions of the input vectors are reduced by the principal components analysis,which can effectively shorten the training time of the network.It can be concluded from the numerical analyses that the improved ART2 neural network can be used to study the new samples itself without destructing the original samples.It can recognize the damage patterns under strong noisy environments rapidly and precisely,and it is also applicable to on-line monitoring.
引文
[1]李宏男.结构健康监测[M].大连:大连理工大学出版社,2004.
[2]李宏男,李东升.土木工程结构安全性评估、健康监测及诊断述评[J].地震工程与工程振动,2002,22(3):82-90.
[3]胡守仁,余少波.戴葵.神经网络导论[M].长沙:国防科技大学出版社,1993.
[4]沈清,汤霖.模式识别导论[M].长沙:国防科技大学出版社,1991.
[5]Carpenter G A,Grossberg S.ART-2:Self-organi-zation of stable category recognition codes for analoginput pattern[J].Applied Optics,1987,26(23):4919-4930.
[6]于秀林,任雪松.多元统计分析[M].北京:中国统计出版社,1999.
[7]杨晓楠,姜绍飞,倪一清,等.基于特征向量简化的复杂土木结构损伤检测[J].沈阳建筑工程学院学报,2003,19(2):81-84.
[8]Johnson E A,Lam H F,Katafygiotis L S and Beck JL.A Benchmark Problem for Structural Health Moni-toring and Damage Detection[A].Proceedings of the14th ASCE Engineering Mechanics Conference[C].USA.:Austin,2000.21-24.
[9]徐永成,温熙森,韩小云.一种新型ARTII无监督分类算法[J].计算机工程与应用,2001,(8):13-15.
[10]黄石生,钱迎雪.基于ART人工神经网络的焊缝跟踪检测算法[J].机械工程学报,1994,30(2):93-98.
[11]Maartje E J,Raijmakers Peter,Molenaar C M.ExactART:A complete implementation of an ART net-work[J].Neural Networks,1997,10(4):649-669.
[12]Tsou P,Shen M H H.Structural damage detectionand identification using neural networks[J].AIAAJournal,1994,32(1):176-183.
[13]Nakamura M,Masri S F,Chassiakos A G,et al.Amethod for non-parametric damage detectionthrough the use of neural networks[J].EarthquakeEngineering&Structural Dynamics,1998,27(9):997-1010.
[2]李宏男,李东升.土木工程结构安全性评估、健康监测及诊断述评[J].地震工程与工程振动,2002,22(3):82-90.
[3]胡守仁,余少波.戴葵.神经网络导论[M].长沙:国防科技大学出版社,1993.
[4]沈清,汤霖.模式识别导论[M].长沙:国防科技大学出版社,1991.
[5]Carpenter G A,Grossberg S.ART-2:Self-organi-zation of stable category recognition codes for analoginput pattern[J].Applied Optics,1987,26(23):4919-4930.
[6]于秀林,任雪松.多元统计分析[M].北京:中国统计出版社,1999.
[7]杨晓楠,姜绍飞,倪一清,等.基于特征向量简化的复杂土木结构损伤检测[J].沈阳建筑工程学院学报,2003,19(2):81-84.
[8]Johnson E A,Lam H F,Katafygiotis L S and Beck JL.A Benchmark Problem for Structural Health Moni-toring and Damage Detection[A].Proceedings of the14th ASCE Engineering Mechanics Conference[C].USA.:Austin,2000.21-24.
[9]徐永成,温熙森,韩小云.一种新型ARTII无监督分类算法[J].计算机工程与应用,2001,(8):13-15.
[10]黄石生,钱迎雪.基于ART人工神经网络的焊缝跟踪检测算法[J].机械工程学报,1994,30(2):93-98.
[11]Maartje E J,Raijmakers Peter,Molenaar C M.ExactART:A complete implementation of an ART net-work[J].Neural Networks,1997,10(4):649-669.
[12]Tsou P,Shen M H H.Structural damage detectionand identification using neural networks[J].AIAAJournal,1994,32(1):176-183.
[13]Nakamura M,Masri S F,Chassiakos A G,et al.Amethod for non-parametric damage detectionthrough the use of neural networks[J].EarthquakeEngineering&Structural Dynamics,1998,27(9):997-1010.
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心