Vous êtes sur la page 1sur 3

The Use Of An Incidence Matrix To Assemble Element Stiffness Matrices M.

Gockel, July 30th, 2010 This technique was presented to me in a class by Prof. DiMaggio of Columbia Univ . in the 1960's. It was taught from his notes, without reference to a text. If anyone knows of any references today I'd appreciate seeing them. It will be de monstrated by example. Given two element stiffness matrices [A] and [B], assemble them into one structu re. Spring A Local Element Local Node Name Node ID ID A 1---------------2 a11 a12 [A]= [ a21 a22 Connect node 2 of A to node 2 of B to produce a structure with nodes 11, 21, and 31. The elements lie along the "1" axis. 1. Assemble the two matrices along the diagonal of a matrix [KELM], in order of increasing local node number. (1) 1) a11 2) a21 KELM = 3)[0 4) 0 (2) a12 a22 0 0 (3) 0 0 b11 b21 (4) 0 0 b12] b22 ], [B] similarly Spring B Local Element Local Node Name Node ID ID B 2---------------1

2. Write a incidence matrix T that maps KELM to KSYS, the system stiffness matr ix, by mapping row and column numbers in KELM to the system nodes 11, 21, 31, wh ich are part of an open set. T*usys = uels Local Node 1 [0 0 0 0 1 0 1 Row in El. matrix

0 u11,1 ua,1 u1 0]*[u21,1] = [ua,2] = [u2] 1 u31,1 ub,1 u3 0 ub,2 u4

[T^t*KELM]*T = [prod]*T = KSYS, where ^t denotes matrix transpose 1 0 0 0 a11 [0 1 0 1]*[a21 0 0 1 0 0 0 a12 a22 0 0 0 0 0 a11 a12 0 0 1 0 ]*T = [a21 a22 b21 b22]*[0 b11 b12 0 0 b11 b12 0 b21 b22 0 0 1 0 1 0 0] 1 0

a11 a12 0 =[a21 (a22+b22) b21] 0 b12 b11

Note that the pre-mulitplication by T^t moves the third and forth rows up to the second and third rows, in a reverse order. The post-multiplication by T sums t he third and fourth columns into the second and third columns, again with a reve rsal. Implementation in NASTRAN The DMIG entry, using the IFO=9 option (rectangular) is a convenient method to i nput T^t, named TTRANS below. In this format the columns are numbered with the sequential numbers that are related to the element node IDs, and the rows are nu mbered with (node ID, C) pairs. The TTRANS matrix input for this model is $ Dmig, Dmig, Dmig, Dmig, Dmig, Notes: 1. The (21,1) row has two entries because both elements attach here. 2. The row numbers use a "C" value of 1 because the input format requires a nonzero value. Any value 1-6 may be used. It is ignored after input. 3. The procedure for filling out this table is: * Make a table that relates element node IDs to the KELM row and column numbers. For example, element B has row 4 at its first end, and row 3 at its second end . * Put the number of DOFs in KELM in the NCOL field of the DMIG entry header. * Number the columns of TTRAMS sequentially from 1. * Put in the system (node ID, C) pair for the non-zero row element in each colum n, with a value of 1.0. TTRANS, TTRANS, TTRANS, TTRANS, TTRANS, 0, 1, 2, 3, 4, IFO 9, , 1, , 1, , 1, , 1, , , , 11, 21, 31, 21, NCOL , 4 1, 1.0 1, 1.0 1, 1.0 1, 1.0

The extension to a full 3-D model with six DOF per node is apparent. * Replace the 1.0 terms in the expression above with an identity matrix of dimen sion six. Account for a different sequence of node variables when the conventio n used in the element matrices differs from that of NASTRAN. * If the element coordinates differ from the basic coordinate system insert a ma trix TBE between T and KELM. [How to do in DMAP?] * If the system coordinate system differs from the basic system insert a matrix TSB before T. * Replace the 1.0 terms in the expression above with an identity matrix of dimen sion six. Both of these Ti coordinate transformation matrices contain 6 by 6 matrices alon g their diagonal. They can be determined by general principles. The data is al so available in the CSTM data block, output by the GP1 module. The data for one coordinate system is contained in one column, so it requires some re-formatting to get it into 6 by 6 matrices, one for each coordinate system. The most effic ient order of operations is likely TALL = TSB*T*TBE KSYS = TALL^t*KELM*TALL

Vous aimerez peut-être aussi