Area Moment Matrices y y' t x' t ey n θ ex ty x θ 1 tx 1 θ nx n ny The transformation equations for area moments due to a rotation of axes are: I x′x ' = I xx cos 2 θ + I yy sin 2 θ − 2 I xy sin θ cos θ I y ' y ' = I xx sin 2 θ + I yy cos 2 θ + 2 I xy sin θ cosθ I x ' y ' = I xx sin θ cos θ − I yy sin θ cosθ + I xy ( cos 2 θ − sin 2 θ ) We can write these relations in terms of the components of the unit vectors n, t acting along the x' and y' axes, respectively, where n = cos θ e x + sin θ e y t = − sin θ e x + cos θ e y where ex , ey are unit vectors along the x and y- axes. We find I x ' x ' = I xx nx2 + I yy n y2 − 2 I xy nx n y I y ' y ' = I xx t x2 + I yy t y2 − 2 I xy t x t y I x ' y ' = − I xx nx t x − I yy n y t y + I xy ( nx t y + n y t x ) These relations can be written in matrix notation as ⎡ Ix'x' ⎢−I ⎣ x' y' or where − I x ' y ' ⎤ ⎡ nx = I y ' y ' ⎥⎦ ⎢⎣ t x n y ⎤ ⎡ I xx t y ⎥⎦ ⎢⎣ − I xy [I '] = [Q] [I ][Q] − I xy ⎤ ⎡ nx I yy ⎥⎦ ⎢⎣ n y tx ⎤ t y ⎥⎦ T ⎡ nx Q = [ ] ⎢n ⎣ y t x ⎤ ⎡ cos ( x, x ') = t y ⎥⎦ ⎢⎣cos ( y, x ') is called the direction cosine matrix cosine of the angle between the x and y' axes, cos ( x, y ') ⎤ etc. ⎥ cos ( y, y ') ⎦ Example (in MATLAB) >> M = [ 288 72;72 72] M= y y' 288 72 72 72 Ixx = 288 in4 Iyy = 72 in4 Ixy = -72 in4 x' 30o x >> angle =30*pi/180; >> Q = [ cos(angle) -sin(angle); sin(angle) cos(angle)] Q= [Q] .. direction cosine matrix 0.8660 -0.5000 0.5000 0.8660 >> Mr =Q'*M*Q [Mr] = [Q]T [M] [Q] Mr = 296.3538 -57.5307 -57.5307 63.6462 Ix'x' = 296 in4 Iy'y' = 64 in4 Ix'y' = 58 in4 It is very easy to determine the principal values and principal directions in this matrix form of the relations Suppose we can find a set of coordinates where the unit vectors along the principal axes have components Nx, Ny and Tx ,Ty . Then ⎡ I1 ⎢0 ⎣ 0 ⎤ ⎡Nx =⎢ ⎥ I 2 ⎦ ⎣ Tx N y ⎤ ⎡ I xx Ty ⎥⎦ ⎢⎣ − I xy − I xy ⎤ ⎡ N x I yy ⎥⎦ ⎢⎣ N y Tx ⎤ Ty ⎥⎦ where I1, I2 are the principal area moments and the mixed area moment is zero in the principal axis coordinates If we multiply both sides of the above equation by [Q] we find, since ⎧1 0 ⎫ T = = Q Q I [ ][ ] [ ] ⎨0 1 ⎬ ⎩ ⎭ ⎡ Nx ⎢N ⎣ y which gives Tx ⎤ ⎡ I1 Ty ⎥⎦ ⎢⎣ 0 0 ⎤ ⎡ I xx =⎢ ⎥ I 2 ⎦ ⎣ − I xy − I xy ⎤ ⎡ N x I yy ⎥⎦ ⎢⎣ N y Tx ⎤ Ty ⎥⎦ ⎡ N x I1 Tx I 2 ⎤ ⎡ I xx ⎢ N I T I ⎥ = ⎢− I y 2⎦ ⎣ y 1 ⎣ xy − I xy ⎤ ⎡ N x I yy ⎥⎦ ⎢⎣ N y Tx ⎤ Ty ⎥⎦ This is equivalent to the two sets of equations: ⎧ N x I1 ⎫ ⎡ I xx − I xy ⎤ ⎧ N x ⎫ ⎨ ⎬=⎢ ⎥⎨ ⎬ N I ⎩ y 1 ⎭ ⎣ − I xy I yy ⎦ ⎩ N y ⎭ ⎧Tx I 2 ⎫ ⎡ I xx − I xy ⎤ ⎧Tx ⎫ ⎨ ⎬=⎢ ⎥ ⎨T ⎬ − T I I I y 2 xy yy ⎩ ⎭ ⎣ ⎦⎩ y⎭ Thus we see to find either the principal area moment I1 and its principal direction N or the principal area moment I2 and its principal direction T we need to solve the system of equations ⎡ I xx ⎢− I ⎣ xy − I xy ⎤ ⎧U x ⎫ ⎧U x I ⎫ ⎨ ⎬=⎨ ⎬ I yy ⎦⎥ ⎩U y ⎭ ⎩U y I ⎭ where the unit vector U can be either N or T and I can be either I1 or I2 The system of equations ⎡ I xx ⎢− I ⎣ xy − I xy ⎤ ⎧U x ⎫ ⎧U x I ⎫ ⎨ ⎬=⎨ ⎬ ⎥ I yy ⎦ ⎩U y ⎭ ⎩U y I ⎭ which can also be written as [I ]{U} = I {U} is called an eigenvalue problem, whose solution is a scalar eigenvalue, I, and a corresponding eigenvector, U Since this eigenvalue problem can be rewritten as ⎡ I xx − I1 ⎢ −I xy ⎣ − I xy ⎤ ⎧U x ⎫ ⎧0 ⎫ ⎨ ⎬=⎨ ⎬ I yy − I ⎥⎦ ⎩U y ⎭ ⎩0 ⎭ to find a solution we must solve the system of equations ( I xx − I )U x − I xyU y = 0 − I xyU x + ( I yy − I )U y = 0 But this is a homogeneous set of equations which only has the solution U = 0 unless the determinant of the matrix of coefficients is zero, i.e. ( I xx − I ) ( I yy − I ) − I xy2 = 0 Expanding this equation we obtain a quadratic equation for I: I 2 − ( I xx + I yy ) I + ( I xx I yy − I xy2 ) = 0 which has the two roots I1 , I 2 I ( = xx + I yy ) 2 ⎛ I xx − I yy ⎞ 2 ± ⎜ ⎟ + I xy 2 ⎠ ⎝ 2 which we see are just the principal area moments ( I xx − I )U x − I xyU y = 0 − I xyU x + ( I yy − I )U y = 0 if we place one of the principal values back into the above system of equations then we can solve for the corresponding principal direction. However, since we have set the determinant of this system equal to zero, the two equations above are not independent. Thus, we can only solve one of them for a ratio of unit vector components. Thus, for example, from the first equation and using I = I1 we have I xy Ux = U y ( I xx − I1 ) Note: this is equivalent to solving for θ via: I xx − I1 ) ( tan θ = I xy But since U is a unit vector we have U x2 + U y2 = 1 which we can solve for Uy in terms of the above ratio as Uy = ±1 (U x / U y ) + 1 2 And then we can find Ux since the ratio Ux/Uy is known. Note that we only get the vector solution to within a plus or minus sign since both U and –U are principal directions. If we repeat the process for I2 then we can find the second principal direction. However, in MATLAB we can get both principal values and directions out directly by just forming up the area moment matrix ⎡ I xx − I xy ⎤ M =⎢ ⎥ I I − xy yy ⎣ ⎦ and then giving that matrix to the built-in function eig which solves the eigenvalue problem. the MATLAB call is: [ pdirs, pvals] = eig(M) The matrix pdirs will then have the principal direction components (in columns) as ⎡(U x )1 pdirs = ⎢ ⎢⎣(U y )1 (U x )2 ⎤ ⎥ (U y )2 ⎥⎦ and the matrix pvals will have the corresponding principal values ⎡ I1 pvals = ⎢ ⎣0 0⎤ I 2 ⎥⎦ I xx = 288 >> M = [ 288 72;72 72]; >> [pdirs, pvals] = eig(M) I yy = 72 I xy = −72 pdirs = 0.2898 -0.9571 Example: -0.9571 -0.2898 pvals = I1 =50.2 U1 = 0.2898 ex - 0.9571ey I2 =309.8 U2 = -0.9571 ex - 0.28981ey angle (degrees) for 309.8 value 50.2002 0 0 309.7998 >> atan(pdirs(2,2)/pdirs(1,2))*180/pi ans = 16.8450 It can be shown that the eigenvalues I1 , I2 of the eigenvalue problem [I ]{U} = I {U} are always real and the eigenvectors U1 , U2 are real and orthogonal to each other since the matrix [ I ] is a real, symmetrical matrix. One of the reasons for treating the transformation of area moments by a matrix approach is that it easily generalizes to more complex problems. For example, in dynamics the three dimensional angular motion of a body (such as a spinning satellite, for example) is controlled by the mass moments of inertia defined as I xxm = ∫ ρ ( y 2 + z 2 ) dV I yym = ∫ ρ ( x 2 + z 2 ) dV I zzm = ∫ ρ ( x 2 + y 2 ) dV I xym = ∫ ρ ( xy ) dV I xzm = ∫ ρ ( xz ) dV I yzm = ∫ ρ ( yz ) dV where ρ is the mass density and dV is a volume element In this case the mass moments transform due to a rotation of axes in just the same manner as we have already discussed. If we let unit vectors n, t, v be along the x', y', z' axes (which are assumed to be orthogonal to each other), then y y' x' t n x v z z' ⎡ Ix'x' ⎢ ⎢− I y ' x ' ⎢−I z 'x' ⎣ −I x' y' I y'y' −Iz' y' − I x ' z ' ⎤ ⎡ nx ⎥ ⎢ − I y ' z ' ⎥ = ⎢ tx I z ' z ' ⎥⎦ ⎢⎣ vx ny ty vy nz ⎤ ⎡ I xx ⎥⎢ t z ⎥ ⎢ − I yx vz ⎥⎦ ⎢⎣ − I zx − I xy I yy − I zy − I xz ⎤ ⎡ nx ⎥ − I yz ⎥ ⎢⎢ n y I zz ⎥⎦ ⎢⎣ nz tx ty tz vx ⎤ v y ⎥⎥ vz ⎥⎦ ⎡ Ix'x' ⎢ ⎢− I y ' x ' ⎢−I z 'x' ⎣ −I x' y' I y'y' −Iz' y' − I x ' z ' ⎤ ⎡ nx ⎥ ⎢ − I y ' z ' ⎥ = ⎢ tx I z ' z ' ⎥⎦ ⎢⎣ vx ny ty vy nz ⎤ ⎡ I xx ⎥⎢ t z ⎥ ⎢ − I yx vz ⎥⎦ ⎢⎣ − I zx − I xy I yy − I zy − I xz ⎤ ⎡ nx ⎥ − I yz ⎥ ⎢⎢ n y I zz ⎥⎦ ⎢⎣ nz tx ty tz vx ⎤ v y ⎥⎥ vz ⎥⎦ We see that again we have [I '] = [Q] [I ][Q] T In this case there are three principal mass moments of inertia and three corresponding principal directions. These are again determined by the solution of the eigenvalue problem [I ]{U} = I {U} Using MATLAB it is still easy to solve for the principal mass moments of inertia and the principal directions with the same eigenvalue function eig >> I=[100 20 30; 20 300 50; 30 50 200] I= 100 20 30 20 300 50 30 mass moment of inertia matrix 50 200 EDU>> [pdirs, pvals]=eig(I) pdirs = 0.9670 -0.2166 0.1337 -0.0322 0.4169 0.9084 -0.2525 -0.8827 0.3962 pvals = 91.4995 0 0 0 183.7468 0 0 0 324.7537 principal directions (in columns) (x, y, z components of a unit vector along the principal axis) principal mass moments of inertia