An enhanced AMLS method and its performance
Introduction
While computation capability has increased rapidly, the demand for large scale finite element (FE) models has increased even more rapidly. Therefore, it has always been an important issue to reduce computational cost. A variety of model reduction methods have been developed and widely used in many engineering fields [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12]. The focus in model reduction is on reducing computational cost with the least possible loss in accuracy.
Within the structural dynamics community, component mode synthesis (CMS) is a popular and effective finite element (FE) model reduction method [5], [6], [7], [8], [9], [10], [11], [12]. In CMS methods, an original (global) FE model is partitioned into smaller substructures, substructural eigenvalue problems are solved, and a reduced model constructed by retaining only dominant substructural modes is used for calculations, instead of the much larger original FE model. For this reason, CMS methods can significantly reduce overall computational cost required for many applications (e.g., controller design for multi-body dynamics systems, structural health monitoring, structural design optimization, model identification).
In the 1990s, the automated multi-level substructuring (AMLS) method, a computer-aided CMS method, was proposed in the field of applied mathematics [13], [14], [15], [16]. Due to its computational efficiency, involving recursive partitioning and matrix reordering processes, the AMLS method has become popular for reduced-order modeling. Recently, Bennighof and Lehoucq [17] proposed a well-defined formulation of the AMLS method based on the concept of the Craig–Bampton (CB) method [6], [17]. The AMLS method has been also used as a solver of eigenvalue problems in many commercial FE software.
In the original CB and AMLS methods, a transformation matrix is constructed by retaining only dominant substructural modes. Using the transformation matrix, original FE models can be transformed into reduced models, which approximate the original models. With this procedure, residual substructural modes are simply truncated without further consideration. However, when the residual mode effect is considered, the accuracy of the original transformation matrix can be improved. That is, the original (global) models can be more precisely approximated. This approach has been used for flexibility based CMS methods, in which, unlike for the CB and AMLS methods, substructures are connected with a free interface [7], [10], [11], [12].
In this study, we derive a new transformation matrix for the AMLS method enhanced by considering the residual mode effect. One difficulty is the fact that the enhanced transformation matrix contains an unknown eigenvalue. In order to approximate the unknown eigenvalue, we adopt O’Callahan’s idea, which was originally proposed to develop the improved reduced system (IRS) method by improving Guyan reduction [18]. Finally, the enhanced transformation matrix is defined without the unknown eigenvalue, and by using the newly defined transformation matrix, an enhanced AMLS method is proposed. The reduced FE models obtained from the enhanced AMLS methods have the same size as those obtained from the original AMLS method. However, compared to the original AMLS method, the enhanced AMLS method can provide significantly improved reduced-order models.
In the following sections, we present the general framework of CMS methods in Section 2, and briefly review the original AMLS method in Section 3. In Section 4, the formulation details of the enhanced AMLS method are presented, and its performance and computational cost are tested in Sections 5 Numerical examples, 6 Computational cost, respectively. The conclusions are given in Section 7.
Section snippets
Component mode synthesis
In this section, the general framework of component mode synthesis (CMS) is briefly presented. In structural dynamics, the linear dynamics equations of a global (non-partitioned) FE model can be expressed as where and are the global mass and stiffness matrices, respectively, and and are the global displacement and force vectors, respectively. Subscript denotes the global structure.
Considering a free harmonic vibration , from Eq. (1), the following eigenvalue
Original AMLS method
Since the AMLS method proposed by Bennighof and his coworkers [17], [19], [20] is based on the Craig–Bampton (CB) method [6], substructures are connected at a fixed interface boundary, see Fig. 1(b). However, unlike for the CB method, the interface boundary DOFs are also considered as substructures in the AMLS method. The interior DOFs are considered as the bottom level substructures and the interface boundary DOFs are considered as the higher level substructures or highest level
Enhanced AMLS method
In the original AMLS method, the reduced transformation matrix can be constructed retaining dominant modes only. However, when residual modes are appropriately considered, the reduced transformation matrix can be enhanced.
The transformation matrix defined in Eq. (5) can be represented as where is the multi-level constraint mode matrix, and is the eigenvector matrix that contains all the
Numerical examples
In this section, we compare the performance of the enhanced AMLS method to the original AMLS method. The original and enhanced AMLS methods were implemented using MATLAB. Four structural problems are considered: rectangular plate, cylindrical solid, bench corner structure and hyperboloid shell problems, in which, for finite element modeling, 4-node MITC shell [23], [24], [25], [26] and 8-node brick finite elements are used.
The frequency cut-off mode selection method is used to select the
Computational cost
In order to investigate the computational cost required for the enhanced AMLS method, computation times are measured, and compared with those of the original AMLS method. A sparse matrix computation with MATLAB is used in a personal computer (Intel core (TM) i7-3770, 3.40 GHz CPU, 16GB RAM). Note that, of course, computation times vary depending on implementation details of the computer codes, as well as on the performance of the computers used. Therefore, the results discussed in this section
Conclusions
In this paper, we presented a new component mode synthesis (CMS) method developed by improving the automated multi-level substructuring (AMLS) method. Unlike for the original AMLS method, the residual mode effect is considered in constructing the transformation matrix. As a result, the original AMLS transformation matrix is enhanced by the residual flexibility, in which the unknown eigenvalue is approximated using O’Callahan’s approach from the improved reduced system (IRS) method.
The enhanced
Acknowledgments
This work was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (No. 2014R1A1A1A05007219), and the Human Resources Development (No. 20134030200300) of the Korea Institute of Energy Technology Evaluation and Planning (KETEP) grant funded by the Korea government Ministry of Trade, Industry and Energy.
References (33)
- et al.
Model reduction using dynamic and iterated IRS techniques
J. Sound Vib.
(1995) Hybrid method of component mode synthesis
Comput. Struct.
(1971)A dual Craig–Bampton method for dynamic substructuring
J. Comput. Appl. Math.
(2004)- et al.
Intrinsic component mode synthesis and plate vibrations
Comput. Struct.
(1992) - et al.
Numerical study of an intrinsic component mode synthesis method
Comput. Methods Appl. Mech. Engrg.
(1992) - et al.
Improving the MITC3 shell finite element by using the Hellinger–Reissner principle
Comput. Struct.
(2012) - et al.
The MITC3+ shell finite element and its performance
Comput. Struct.
(2014) - et al.
Estimating relative eigenvalue errors in the Craig–Bampton method
Comput. Struct.
(2014) - et al.
An accurate error estimator for Guyan reduction
Comput. Methods Appl. Mech. Engrg.
(2014) - et al.
Nonlinear dynamics of flexible multibody systems
Comput. Struct.
(2003)
Reduction of stiffness and mass matrices
AIAA J.
Structural eigenvalue problems: elimination of unwanted variables
AIAA J.
A new iterative order reduction (IOR) method for eigensolutions of large structures
Internat. J. Numer. Methods Engrg.
Dynamic analysis of structural systems using component modes
AIAA J.
Coupling of substructures for dynamic analysis
AIAA J.
Vibration analysis of structures by component mode substitution
AIAA J.
Cited by (49)
Hybrid vibro-acoustic model reduction for model updating in nuclear power plant pipeline with undetermined boundary conditions
2024, Nuclear Engineering and TechnologyDynamic reanalysis of structures with geometric variability and parametric uncertainties via an adaptive model reduction method
2023, Mechanical Systems and Signal ProcessingRefining characteristic constraint modes of component mode synthesis with residual modal flexibility
2022, Mechanical Systems and Signal Processing3D-based equivalent model of SMART control rod drive mechanism using dynamic condensation method
2022, Nuclear Engineering and TechnologyImproving the generalized Bloch mode synthesis method using algebraic condensation
2021, Computer Methods in Applied Mechanics and Engineering