def setup(self): self.surface = surface = self.options['surface'] mesh = surface['mesh'] nx = self.nx = mesh.shape[0] ny = self.ny = mesh.shape[1] self.add_input('mesh', val=np.ones((self.nx, self.ny, 3)), units='m') self.add_input('disp', val=np.ones((self.ny, 6)), units='m') self.add_input('transformation_matrix', val=np.ones((self.ny, 3, 3))) self.add_input('nodes', val=np.ones((self.ny, 3)), units='m') self.add_output('def_mesh', val=np.random.random_sample((self.nx, self.ny, 3)), units='m') # Create index arrays for each relevant input and output. # This allows us to set up the rows and cols for the sparse Jacobians. disp_indices = get_array_indices(self.ny, 6) nodes_indices = get_array_indices(self.ny, 3) mesh_indices = get_array_indices(self.nx, self.ny, 3) transform_indices = get_array_indices(self.ny, 3, 3) mesh_disp_indices = get_array_indices(self.nx, self.ny, 3) # Set up the rows and cols for `def_mesh` wrt `disp` rows = mesh_disp_indices.flatten() cols = np.einsum('i,jk->ijk', np.ones(self.nx), disp_indices[:, :3]).flatten() self.declare_partials('def_mesh', 'disp', val=1., rows=rows, cols=cols) # Set up the rows and cols for `def_mesh` wrt `nodes` rows = np.einsum('ijk,l->ijkl', mesh_disp_indices, np.ones(3, int)).flatten() cols = np.einsum('ik,jl->ijkl', np.ones((self.nx, 3), int), nodes_indices).flatten() self.declare_partials('def_mesh', 'nodes', rows=rows, cols=cols) # Set up the rows and cols for `def_mesh` wrt `mesh` rows = np.einsum('ijk,l->ijkl', mesh_disp_indices, np.ones(3, int)).flatten() cols = np.einsum('ijl,k->ijkl', mesh_indices, np.ones(3, int)).flatten() self.declare_partials('def_mesh', 'mesh', rows=rows, cols=cols) # Set up the rows and cols for `def_mesh` wrt `transformation_matrix` rows = np.einsum('ijl,k->ijkl', mesh_disp_indices, np.ones(3, int)).flatten() cols = np.einsum('jlk,i->ijkl', transform_indices, np.ones(self.nx, int)).flatten() self.declare_partials('def_mesh', 'transformation_matrix', rows=rows, cols=cols)
def setup(self): self.surface = surface = self.options['surface'] mesh=surface['mesh'] nx = self.nx = mesh.shape[0] ny = self.ny = mesh.shape[1] self.add_input('mesh', val=np.ones((self.nx, self.ny, 3)), units='m') self.add_input('disp', val=np.ones((self.ny, 6)), units='m') self.add_input('transformation_matrix', val=np.ones((self.ny, 3, 3))) self.add_input('nodes', val=np.ones((self.ny, 3)), units='m') self.add_output('def_mesh', val=np.random.random_sample((self.nx, self.ny, 3)), units='m') # Create index arrays for each relevant input and output. # This allows us to set up the rows and cols for the sparse Jacobians. disp_indices = get_array_indices(self.ny, 6) nodes_indices = get_array_indices(self.ny, 3) mesh_indices = get_array_indices(self.nx, self.ny, 3) transform_indices = get_array_indices(self.ny, 3, 3) mesh_disp_indices = get_array_indices(self.nx, self.ny, 3) # Set up the rows and cols for `def_mesh` wrt `disp` rows = mesh_disp_indices.flatten() cols = np.einsum('i,jk->ijk', np.ones(self.nx), disp_indices[:, :3]).flatten() self.declare_partials('def_mesh', 'disp', val=1., rows=rows, cols=cols) # Set up the rows and cols for `def_mesh` wrt `nodes` rows = np.einsum('ijk,l->ijkl', mesh_disp_indices, np.ones(3, int)).flatten() cols = np.einsum('ik,jl->ijkl', np.ones((self.nx, 3), int), nodes_indices).flatten() self.declare_partials('def_mesh', 'nodes', rows=rows, cols=cols) # Set up the rows and cols for `def_mesh` wrt `mesh` rows = np.einsum('ijk,l->ijkl', mesh_disp_indices, np.ones(3, int)).flatten() cols = np.einsum('ijl,k->ijkl', mesh_indices, np.ones(3, int)).flatten() self.declare_partials('def_mesh', 'mesh', rows=rows, cols=cols) # Set up the rows and cols for `def_mesh` wrt `transformation_matrix` rows = np.einsum('ijl,k->ijkl', mesh_disp_indices, np.ones(3, int)).flatten() cols = np.einsum('jlk,i->ijkl', transform_indices, np.ones(self.nx, int)).flatten() self.declare_partials('def_mesh', 'transformation_matrix', rows=rows, cols=cols)
def setup(self): self.surface = surface = self.options['surface'] mesh=surface['mesh'] self.nx = mesh.shape[0] self.ny = mesh.shape[1] self.add_input('disp', val=np.zeros((self.ny, 6)), units='m') self.add_output('transformation_matrix', shape=(self.ny, 3, 3)) # Create index arrays for each relevant input and output. # This allows us to set up the rows and cols for the sparse Jacobians. disp_indices = get_array_indices(self.ny, 6) transform_indices = get_array_indices(self.ny, 3, 3) # Set up the rows and cols for `transformation_matrix` wrt `disp` rows = np.einsum('ijk,l->ijkl', transform_indices, np.ones(3, int)).flatten() cols = np.einsum('il,jk->ijkl', get_array_indices(self.ny, 6)[:, 3:], np.ones((3, 3), int)).flatten() self.declare_partials('transformation_matrix', 'disp', rows=rows, cols=cols)
def setup(self): self.surface = surface = self.options['surface'] mesh=surface['mesh'] self.nx = mesh.shape[0] self.ny = mesh.shape[1] self.add_input('disp', val=np.zeros((self.ny, 6)), units='m') self.add_output('transformation_matrix', shape=(self.ny, 3, 3)) # Create index arrays for each relevant input and output. # This allows us to set up the rows and cols for the sparse Jacobians. disp_indices = get_array_indices(self.ny, 6) transform_indices = get_array_indices(self.ny, 3, 3) # Set up the rows and cols for `transformation_matrix` wrt `disp` rows = np.einsum('ijk,l->ijkl', transform_indices, np.ones(3, int)).flatten() cols = np.einsum('il,jk->ijkl', get_array_indices(self.ny, 6)[:, 3:], np.ones((3, 3), int)).flatten() self.declare_partials('transformation_matrix', 'disp', rows=rows, cols=cols)