wawi 0.0.1__py3-none-any.whl → 0.0.3__py3-none-any.whl

wawi/modal.py

@@ -0,0 +1,608 @@
+import numpy as np
+from .tools import print_progress as pp
+from scipy.interpolate import interp1d
+
+def maxreal(phi):
+    """
+    Rotate complex vectors (stacked column-wise) such that the absolute values of the real parts are maximized.
+
+    Arguments
+    ---------------------------
+    phi : double
+        complex-valued modal transformation matrix (column-wise stacked mode shapes)
+
+    Returns
+    ---------------------------
+    phi_max_real : boolean
+        complex-valued modal transformation matrix, with vectors rotated to have maximum real parts
+    """   
+
+    angles = np.expand_dims(np.arange(0,np.pi/2, 0.01), axis=0)
+    phi_max_real = np.zeros(np.shape(phi)).astype('complex')
+    for mode in range(0,np.shape(phi)[1]):
+        rot_mode = np.dot(np.expand_dims(phi[:, mode], axis=1), np.exp(angles*1j))
+        max_angle_ix = np.argmax(np.sum(np.real(rot_mode)**2,axis=0), axis=0)
+
+        phi_max_real[:, mode] = phi[:, mode] * np.exp(angles[0, max_angle_ix]*1j)*np.sign(sum(np.real(phi[:, mode])))
+
+    return phi_max_real
+
+def get_mode_sort(lambd, order_fun_dict=None, return_as_dict=True, remove_conjugates=['dynamic'], sort=['dynamic']):
+    order_fun_dict_std = {
+        'dynamic': lambda l: np.logical_and(np.imag(l)!=0, np.real(l)<0),
+        'undamped':lambda l: np.real(l)==0,
+        'unstable': lambda l: np.logical_and(np.imag(l)!=0, np.real(l)>0),
+        'overdamped': lambda l: np.logical_and(np.imag(l)==0, np.real(l)<0)
+        }
+        
+    if order_fun_dict is None:
+        order_fun_dict = order_fun_dict_std  
+    elif type(order_fun_dict) is list:  #specify certain terms from standard dict defined
+        order_fun_dict = {key:order_fun_dict_std[key] for key in order_fun_dict}
+        
+    ix_dict = dict()    
+    
+    for key, fun in order_fun_dict.items():     
+        ix = np.where(fun(lambd))[0]
+        
+        if key in sort:
+            ix2 = np.argsort(np.abs(lambd[ix]))
+            ix = ix[ix2]
+            
+        if remove_conjugates:
+            ix = ix[::2]
+            
+        ix_dict[key] = ix
+        
+        
+    if return_as_dict:
+        return ix_dict
+    else:
+        return np.hstack(list(ix_dict.values()))
+       
+def sort_modes(lambd, phi=None, order_fun_dict=None, return_as_dict=True, remove_conjugates=['dynamic'], sort=['dynamic']):
+    ix_dict = get_mode_sort(lambd, order_fun_dict=order_fun_dict, return_as_dict=True, remove_conjugates=['dynamic'], sort=['dynamic'])
+    lambd_dict = dict()
+    
+    if phi is not None:
+        phi_dict = dict()
+
+    for key in ix_dict:
+        ix = ix_dict[key]
+        lambd_dict[key] = lambd[ix]
+        if phi is not None:
+            phi_dict[key] = phi[:, ix]       
+
+    if not return_as_dict:
+        lambd_list = [lambdi for lambdi in list(lambd_dict.values()) if lambdi.size!=0]
+        lambd_sorted = np.hstack([lambd_list])
+        
+        if phi is not None:
+            phi_list = [phii for phii in list(phi_dict.values()) if phii.size!=0]
+            phi_sorted = np.hstack(phi_list)
+        else:
+            phi_sorted = None
+        return lambd_sorted, phi_sorted
+    else:
+        return lambd_dict, phi_dict
+        
+
+def statespace(K, C, M):
+    ndofs = np.shape(K)[0]
+    A = np.zeros([2*ndofs, 2*ndofs])
+    A[0:ndofs, ndofs:2*ndofs] = np.eye(ndofs)
+    A[ndofs:2*ndofs, 0:ndofs] = -np.linalg.inv(M) @ K
+    A[ndofs:2*ndofs, ndofs:2*ndofs] = -np.linalg.inv(M) @ C
+
+    return A
+
+
+def iteig(K, C, M, omega=None, omega_ref=0, input_functions=True, itmax=None, tol=None, keep_full=False,
+          mac_min=0.9, w_initial=None, normalize=False, print_progress=False, print_warnings=True,
+          track_by_psi=True, remove_velocity=True, divergence_protection=True):  
+    
+    mean_w = False
+    
+    if itmax is None:
+        itmax = 15
+    
+    if tol is None:
+        tol = 1e-4
+        
+    
+    if not input_functions:
+        K = interp1d(omega, K, kind='quadratic', fill_value='extrapolate')
+        C = interp1d(omega, C, kind='quadratic', fill_value='extrapolate')
+        M = interp1d(omega, M, kind='quadratic', fill_value='extrapolate')
+     
+    ndofs = K(0).shape[0]
+    lambd = np.zeros([2*ndofs], dtype=complex)
+    q = np.zeros([2*ndofs, 2*ndofs], dtype=complex)
+       
+    # Reference phi
+    A = statespace(K(omega_ref), C(omega_ref), M(omega_ref))
+    lambd_ref, q_ref = np.linalg.eig(A)
+    not_converged = []
+    omega_ref = np.abs(np.imag(lambd_ref))
+    
+    if w_initial is None:
+        w_initial = np.sort(np.abs(np.imag(lambd_ref)))
+
+    for m in range(2*ndofs):
+        w = w_initial[m]
+        
+        wprev = np.inf
+        wprev2 = -np.inf
+        
+        phi = q[:,0]*0   
+        phiprev = 0*phi
+        phiprev2 = 0*phi
+        
+        q_ref_m = q_ref[:, m:m+1] 
+
+        for i in range(0, itmax):     
+            A = statespace(K(w), C(w), M(w))
+            lambdai, qi = np.linalg.eig(A)
+            
+            if track_by_psi:
+                macs = xmacmat(qi, phi2=q_ref_m, conjugates=False)
+                m_ix = np.argmax(macs, axis=0)[0]
+                
+                phi = qi[:, m_ix:m_ix+1]
+                lambdai = lambdai[m_ix]
+            else:
+                phi = qi[:, m:m+1]
+                lambdai = lambdai[m]
+
+            w = abs(np.imag(lambdai))
+            if mean_w:
+                w = (wprev+w)/2
+            
+            if w==0:
+                mean_w = False
+                break
+            elif np.abs(w - wprev2)/w <= tol and np.abs(w - wprev)/w <= tol and mac(phi, phiprev)>=mac_min and mac(phi, phiprev2)>=mac_min:    # Converged!
+                mean_w = False    
+                break
+            elif np.abs(w - wprev2)/w <= tol and np.abs(w - wprev)/w > tol and mac(phi, phiprev)<mac_min and mac(phi, phiprev2)>=mac_min:  
+                if divergence_protection:
+                    mean_w = True
+                    extra_message = 'Divergence protection active - attempting to continue with average values'
+                else:
+                    extra_message = ''
+                if print_warnings:
+                    print(f'Warning: Suspecting oscillatory divergence on iteration of mode {int(np.ceil(m/2))}. Please conduct checks.')
+                    print(f'Re(lambd) = {np.real(lambdai):.2f}, Im(lambd) = {np.imag(lambdai):.2f}%')
+                    print(extra_message)
+                
+            wprev2 = wprev*1
+            wprev = w*1
+            
+            phiprev2 = phiprev*1
+            phiprev = phi*1
+    
+            if i is itmax-1:
+                if print_warnings:
+                    print(f'Warning: Maximum number of iterations ({i+1}) performed on mode {int(np.ceil(m/2))}. Storing last value.')
+                    print(f'Last three damped natural frequencies: {wprev2:.2f}, {wprev:.2f}, {w:.2f} rad/s')
+                    print(f'Corresponding automacs (3,1) and (3,2): {xmacmat(phi, phiprev2):.1f}, {xmacmat(phi, phiprev):.1f}')
+                not_converged.append(int(m/2))
+        
+        lambd[m] = lambdai
+        q[:, m] = phi[:,0]
+
+        if print_progress:
+            pp(m+2,2*ndofs, sym='>', postfix=' finished with iterative modal analysis.')
+
+    if print_progress:
+        print(' ')
+    
+    if remove_velocity:
+        q = q[:ndofs, :]  
+        
+    if not keep_full:
+        lambd = lambd[::2]
+        q = q[:, ::2]
+    
+    if normalize==True:
+        for mode in range(0, np.shape(q)[1]):
+            q[:, mode] = q[:, mode]/max(abs(q[:, mode]))
+            
+    # Sort 
+    ix = np.argsort(np.abs(lambd))
+    lambd = lambd[ix]
+    q = q[:,ix]
+    
+    return lambd, q, not_converged
+
+
+def iteig_naive(K, C, M, itmax=None, tol=1e-4):  
+    
+    if itmax is None:
+        itmax = 15
+    
+    if tol is None:
+        tol = 1e-4
+            
+    ndofs = K(0).shape[0]
+    lambd = np.zeros([2*ndofs], dtype=complex)
+    q = np.zeros([2*ndofs, 2*ndofs], dtype=complex)
+       
+    for m in range(0, 2*ndofs, 2):
+        w = 0.0
+        wprev = np.inf
+
+        for i in range(0, itmax):     
+            A = statespace(K(w), C(w), M(w))
+            lambdai, qi = np.linalg.eig(A)
+            ix = np.argsort(np.abs(lambdai))
+            lambdai = lambdai[ix][m]
+            qi = qi[:, ix][:, m]
+ 
+            w = abs(-np.imag(lambdai))
+
+            if np.abs(w - wprev) <= tol:
+                break
+                
+            wprev = w*1
+        
+        lambd[m] = lambdai
+        q[:, m] = qi
+
+    return lambd, q
+
+
+def iteig_freq(K, C, M, omega=None, itmax=15, reference_omega=0, input_functions=True, 
+               tol=1e-4, keep_full=False, mac_min=0.98, w_initial=None, 
+               normalize=False, print_progress=False, print_warnings=True, divergence_protection=True):  
+    
+    if not input_functions:
+        K = interp1d(omega, K, kind='quadratic', fill_value='extrapolate')
+        C = interp1d(omega, C, kind='quadratic', fill_value='extrapolate')
+        M = interp1d(omega, M, kind='quadratic', fill_value='extrapolate')
+        
+    if divergence_protection:
+        extra_message = '[Divergence protection active]'
+    else:
+        extra_message = ''
+        
+    ndofs = K(0).shape[0]
+
+    lambd = np.zeros([2*ndofs], dtype=complex)
+    q = np.zeros([2*ndofs, 2*ndofs], dtype=complex)
+     
+    not_converged = []
+    
+    if not w_initial:
+        w_initial = np.zeros([ndofs])
+    
+    for m in range(0, 2*ndofs):
+        w = w_initial[int(m/2)]
+        
+        wprev = np.inf
+        wprev2 = -np.inf
+        
+        phi = q[:,0]*0   
+        phiprev = 0*phi
+        phiprev2 = 0*phi
+
+        for i in range(0, itmax):          
+            A = statespace(K(w), C(w), M(w))
+            
+            lambdai, qi = np.linalg.eig(A)
+            sortix = np.argsort(np.imag(lambdai))
+            
+            qi = qi[:, sortix]
+            lambdai = lambdai[sortix]
+            
+            w = abs(np.imag(lambdai[m]))
+            phi = qi[:, m]
+
+            if np.abs(w - wprev2) <= tol and np.abs(w - wprev) <= tol and mac(phi, phiprev)>=mac_min and mac(phi, phiprev2)>=mac_min:    # Converged!
+                break
+            elif np.abs(w - wprev2) <= tol and np.abs(w - wprev) > tol and mac(phi, phiprev)<mac_min and mac(phi, phiprev2)>=mac_min:  
+                if divergence_protection:
+                    w = (w+wprev)/2
+                if print_warnings:
+                    print(f'Oscillatory divergence, mode {np.ceil(m/2):.0f} omega = {w:.2f}, xi = {100*-np.real(lambdai[m])/np.abs(lambdai[m]):.1f} % {extra_message}')
+                    
+            wprev2 = wprev*1
+            wprev = w*1
+            
+            phiprev2 = phiprev*1
+            phiprev = phi*1
+    
+            if i is itmax-1:
+                if print_warnings:
+                    print('** Maximum number of iterations (%i) performed on mode %i. Storing last value.' % (i+1, np.ceil(m/2)))
+                    # print('Last three damped natural frequencies: %f, %f, %f rad/s' % (wprev2, wprev, w))
+                    # print('Corresponding automacs (3,1) and (3,2): %f, %f' % (xmacmat(phi, phiprev2), xmacmat(phi, phiprev)))
+                not_converged.append(int(m/2))
+                
+        lambd[m] = lambdai[m]
+        q[:, m] = qi[:, m] 
+
+        if print_progress:
+            pp(m+2,2*ndofs, sym='>', postfix=' finished with iterative modal analysis.')
+
+    if print_progress:
+        print(' ')
+        
+    if not keep_full:
+        q = q[0:ndofs, :]    
+        lambd = lambd[0::2]
+        q = q[:, 0::2]
+    else:
+        lambd[1::2] = np.conj(lambd[0::2])
+        q[:, 1::2] = np.conj(q[:,0::2])
+    
+    if normalize==True:
+        for mode in range(0, np.shape(q)[1]):
+            q[:, mode] = q[:, mode]/max(abs(q[:, mode]))
+
+    # Sort 
+    ix = np.argsort(np.abs(lambd))
+    lambd = lambd[ix]
+    q = q[:,ix]
+
+    return lambd, q, not_converged
+
+    
+
+def xmacmat(phi1, phi2=None, conjugates=True):
+    """
+    Modal assurance criterion numbers, cross-matrix between two modal transformation matrices (modes stacked as columns).
+
+    Arguments
+    ---------------------------
+    phi1 : double
+        reference modes
+    phi2 : double, optional
+        modes to compare with, if not given (i.e., equal default value None), phi1 vs phi1 is assumed
+    conjugates : True, optional
+        check the complex conjugates of all modes as well (should normally be True)
+
+    Returns
+    ---------------------------
+    macs : double
+        matrix of MAC numbers
+    """
+    # If no phi2 is given, assign value of phi1
+    if phi2 is None:
+        phi2 = 1.0*phi1
+        
+    if len(np.shape(phi1))==1:
+        phi1 = np.expand_dims(phi1, axis=0).T
+        
+    if len(np.shape(phi2))==1:
+        phi2 = np.expand_dims(phi2, axis=0).T
+
+    # norms1 = np.dot(np.expand_dims(np.sum(phi1.T * phi1.T.conj(), axis=1), axis=0), np.expand_dims(np.sum(phi2.T * phi2.T.conj(),axis=1), axis=1))
+    norms = np.real(np.sum(phi1.T * np.conj(phi1.T), axis=1))[:,np.newaxis] @ np.real(np.sum(phi2.T * np.conj(phi2.T),axis=1))[np.newaxis,:]
+
+
+    if conjugates:
+        macs1 = np.divide(abs(np.dot(phi1.T, phi2))**2, norms)
+        macs2 = np.divide(abs(np.dot(phi1.T, phi2.conj()))**2, norms)     
+        macs = np.maximum(macs1, macs2)
+    else:
+        macs = np.divide(abs(np.dot(phi1.T, phi2))**2, norms)
+
+    macs = np.real(macs)
+    
+    if np.size(macs) == 1:
+        macs = macs[0,0]
+    
+    return macs
+
+def mac(phi1, phi2):
+
+    mac_value = np.real(np.abs(np.dot(phi1.T,phi2))**2 / np.abs((np.dot(phi1.T, phi1) * np.dot(phi2.T, phi2))))
+    return mac_value
+
+
+
+def mcf(phi):
+
+    # Ensure on matrix format
+    if phi.ndim == 1:
+        phi = phi[:,np.newaxis]
+
+
+    n_modes = np.shape(phi)[1]
+
+    X = np.real(phi)
+    Y = np.imag(phi)
+
+    modal_complexity_factor = [None]*n_modes
+    for mode in range(0,n_modes):
+        modal_complexity_factor[mode] = np.abs(np.dot(X[:,mode], Y[:,mode]))**2 / (np.abs(np.dot(X[:,mode], X[:,mode])) * np.abs(np.dot(Y[:,mode], Y[:,mode])))
+
+    modal_complexity_factor = np.array(modal_complexity_factor)
+    return modal_complexity_factor
+
+
+def mpc(phi):
+    # Based on the current paper:
+    # Pappa, R. S., Elliott, K. B., & Schenk, A. (1993). 
+    # Consistent-mode indicator for the eigensystem realization algorithm. 
+    # Journal of Guidance, Control, and Dynamics, 16(5), 852–858.
+
+    # Ensure on matrix format
+    if phi.ndim == 1:
+        phi = phi[:,np.newaxis]
+
+    n_modes = np.shape(phi)[1]
+    mpc_val = [None]*n_modes
+
+    for mode in range(0,n_modes):
+        phin = phi[:, mode]
+        Sxx = np.dot(np.real(phin), np.real(phin))
+        Syy = np.dot(np.imag(phin), np.imag(phin))
+        Sxy = np.dot(np.real(phin), np.imag(phin))
+
+        eta = (Syy-Sxx)/(2*Sxy)
+
+        lambda1 = (Sxx+Syy)/2 + Sxy*np.sqrt(eta**2+1)
+        lambda2 = (Sxx+Syy)/2 - Sxy*np.sqrt(eta**2+1)
+
+        mpc_val[mode] = ((lambda1-lambda2)/(lambda1+lambda2))**2
+
+    mpc_val = np.array(mpc_val)
+    return mpc_val
+
+
+def scale_phi(phi, scaling):
+    phi_scaled = phi*1
+    for mode in range(phi.shape[1]):
+        phi_scaled[:,mode] = phi[:,mode] * scaling[mode]
+        
+    return phi_scaled
+
+def normalize_phi(phi, include_dofs=[0,1,2,3,4,5,6], n_dofs=6):
+    phi_n = phi*0
+
+    phi_for_scaling = np.vstack([phi[dof::n_dofs, :] for dof in include_dofs])
+    mode_scaling = np.max(np.abs(phi_for_scaling), axis=0)
+    ix_max = np.argmax(np.abs(phi_for_scaling), axis=0)
+    signs = np.sign(phi_for_scaling[ix_max, range(0, len(ix_max))])
+    signs[signs==0] = 1
+    mode_scaling[mode_scaling==0] = 1
+
+    phi_n = phi/np.tile(mode_scaling[np.newaxis,:]/signs[np.newaxis,:], [phi.shape[0], 1])
+
+    return phi_n, mode_scaling
+
+
+def restructure_as_ref(phi_ref, phi, min_mac=0.0, ensure_unique=True, 
+                       return_all=True, accept_conjugates=True):
+    """
+    Restructure modes based on reference modal transformation matrix phi_ref.
+
+    Arguments
+    ---------------------------
+    phi_ref : double
+        reference modes
+    phi : double
+        modes to compare with
+    min_mac : double, 0.0
+        minimum MAC value for mode to be placed in index arrays
+
+    Returns
+    ---------------------------
+    ixs : integers
+        array of indices, describing which elements to copy to reference 
+        numbering (indices indicate position in original numbering, 
+                   as in phi - position relates to reference numbering)
+    ref_ixs : integers
+        array of indices, describing which elements to assign values to 
+        (using reference numbering) - equal to 0:n_modes if min_mac=0.0 (all are populated).
+    """
+    xmacs = xmacmat(phi_ref, phi2=phi, conjugates=accept_conjugates)  # n_modes_ref-by-n_modes
+    ixs = np.argmax(xmacs, axis=1)      # indices of what mode in phi_ref each mode in phi2 are closest to, i.e, phi_rs = phi[:, ix]
+    ref_ixs = np.arange(len(ixs))
+    xmacs_out = xmacs*1
+
+    if ensure_unique:
+        for row in range(xmacs.shape[0]):
+            xmac_row = xmacs[row, :]
+            xmacs[row, xmac_row<np.max(xmac_row)] = 0
+            
+        for col in range(xmacs.shape[1]):
+            xmac_col = xmacs[:, col]
+            xmacs[xmac_col<np.max(xmac_col), col] = 0
+
+    ok = np.diag(xmacs[:, ixs])>min_mac
+
+    ixs = ixs[ok]
+    ref_ixs = ref_ixs[ok]        
+    rejected_ixs = np.where(~ok)[0]
+    xmacs_out = xmacs_out[ref_ixs, ixs]
+
+    if return_all:
+        return ixs, ref_ixs, rejected_ixs, xmacs_out
+    else:
+        return ixs
+
+
+def robust_phi_restructure(phi_ref, phi, accept_conjugates=False):
+    """
+    Restructure modes based on reference modal transformation matrix phi_ref.
+
+    Arguments
+    ---------------------------
+    phi_ref : double
+        reference modes
+    phi : double
+        modes to compare with
+
+    Returns
+    ---------------------------
+    ixs : integers
+        array of indices, describing which elements to copy to reference numbering (indices indicate position in original numbering, as in phi - position relates to reference numbering)
+    discarded_ixs : integers
+        rest of modes (if not the best match to any mode in reference mode)
+   """
+   
+    xmacs = xmacmat(phi_ref, phi2=phi, conjugates=accept_conjugates)  # n_modes_ref-by-n_modes
+    ixs = np.argmax(xmacs, axis=0)   
+
+    # Make unique based on selected columns in ixs
+    for ix in np.unique(ixs):
+        all_this_ix_ix = np.where(ixs==ix)[0]
+        ix_max = np.argmax(xmacs[:, ixs==ix], axis=0)
+        
+        n = len(all_this_ix_ix)
+        discards = np.array([i for i in range(n) if i != ix_max])
+
+        if len(discards)>=0:
+            ixs[all_this_ix_ix[discards]] = np.nan
+            
+    all_ixs = np.arange(0, phi_ref.shape[1])
+    discarded_ixs = np.sort([i for i in all_ixs if i not in ixs])
+
+    return ixs, discarded_ixs
+
+
+
+def freq_eig(K, C, M, omega, omega_ref=None, phi_ref=None, min_mac=0.0, input_functions=True, keep_full=False):
+    if not input_functions:
+        K = interp1d(omega, K, kind='quadratic', fill_value='extrapolate')
+        C = interp1d(omega, C, kind='quadratic', fill_value='extrapolate')
+        M = interp1d(omega, M, kind='quadratic', fill_value='extrapolate')
+        
+    if phi_ref is None:
+        if omega_ref is None:
+            prev_compare = True
+            omega_ref = omega[0]
+        else:
+            prev_compare = False
+            
+        A_ref = statespace(K(omega_ref), C(omega_ref), M(omega_ref))
+        lambda_ref, phi_ref = np.linalg.eig(A_ref)
+        sortix = np.argsort(abs(lambda_ref))
+        phi_ref = phi_ref[:, sortix]
+    else:
+        prev_compare = False            
+    
+    phi = [None]*len(omega)
+    lambd = [None]*len(omega)
+
+    for k, omega_k in enumerate(omega):
+        phi[k] = np.nan*phi_ref
+        lambd[k] = np.nan*np.ones(phi_ref.shape[1]).astype('complex')
+        
+        A = statespace(K(omega_k), C(omega_k), M(omega_k))
+        lambd_k, phi_k = np.linalg.eig(A)
+        mode_ix, ref_ix, __, __ = restructure_as_ref(phi_ref, phi_k, min_mac=min_mac, ensure_unique=False)
+        lambd[k][ref_ix] = lambd_k[mode_ix]
+        phi[k][:, ref_ix] = phi_k[:, mode_ix]
+        
+        if prev_compare:
+            phi_ref = phi[k]
+            
+    if not keep_full:
+        lambd = [lambd_k[::2] for lambd_k in lambd]
+        phi = [phi_k[::2,::2] for phi_k in phi]
+        
+    return lambd, phi