基于人工神经网络的复杂结构模态匹配  

ARTIFICIAL NEURAL NETWORK BASED METHOD FOR MODE VECTOR CORRELATION

作　　者：汤凯[1] 刘济科[1] 

机构地区：[1]中山大学应用力学与工程系,广州510275

出　　处：《振动与冲击》2006年第1期42-46,共5页Journal of Vibration and Shock

基　　金：高等学校优秀青年教师教学科研奖励计划资助;国家自然科学基金(10102023);广东省自然科学基金(031588;001180);广东省科技计划资助项目(2002C32405)等的资助

摘　　要：对结构模态匹配方法进行简单评述,并提出了一种基于人工神经网络的复杂结构模态匹配方法。该方法能够准确匹配复杂结构模态向量,同时可以有效消除由于试验模态测量中测点不足、测点布置不当、测量不准确等因素导致的试验模态向量间的空间混叠影响。考虑实际工程中能够获得的训练样本数目有限,采用了轮换式学习方法有效扩展了学习样本集。算例表明,本方法合理可靠,在实际工程中有广泛的应用前景。The problem of matching modal vectors that stem from two different sources is addressed and a simple(review) is made.The traditional correlation methods using modal properties are based on MAC(Modal Assurance(Criterion)),including the CMPs(Correlated Mode Pairs),the auto-MAC plot,the COMAC(Coordinate MAC),the FMAC(MAC plot with Frequency scales),and the improved MAC method.These methods are frequently used and proved to be effective.The proposed method in the present paper uses the pattern recognition property of ANNs(Artificial Neural Networks) to solve the problem.A finite element model of a 12-story steel frame is used to illustrate the(performance) of both the traditional and the presented strategies and satisfactory results are observed especially its ability of eliminating spacial aliasing.

关 键 词：模态匹配 人工神经网络 空间混叠 轮换式学习方法 

分 类 号：O32[理学—一般力学与力学基础] TU317[理学—力学]