def solve_multipliers(self, op_point=None, sol_type='dict'): """Solves for the values of the lagrange multipliers symbolically at the specified operating point Parameters ---------- op_point : dict or iterable of dicts, optional Point at which to solve at. The operating point is specified as a dictionary or iterable of dictionaries of {symbol: value}. The value may be numeric or symbolic itself. sol_type : str, optional Solution return type. Valid options are: - 'dict': A dict of {symbol : value} (default) - 'Matrix': An ordered column matrix of the solution """ # Determine number of multipliers k = len(self.lam_vec) if k == 0: raise ValueError("System has no lagrange multipliers to solve for.") # Compose dict of operating conditions if isinstance(op_point, dict): op_point_dict = op_point elif isinstance(op_point, collections.Iterable): op_point_dict = {} for op in op_point: op_point_dict.update(op) else: op_point_dict = {} # Compose the system to be solved mass_matrix = self.mass_matrix.col_join((-self.lam_coeffs.row_join( zeros(k, k)))) force_matrix = self.forcing.col_join(self._f_cd) # Sub in the operating point mass_matrix = _subs_keep_derivs(mass_matrix, op_point_dict) force_matrix = _subs_keep_derivs(force_matrix, op_point_dict) # Solve for the multipliers sol_list = mass_matrix.LUsolve(-force_matrix)[-k:] if sol_type == 'dict': return dict(zip(self.lam_vec, sol_list)) elif sol_type == 'Matrix': return Matrix(sol_list) else: raise ValueError("Unknown sol_type {:}.".format(sol_type))
def to_linearizer(self): """Returns an instance of the Linearizer class, initiated from the data in the KanesMethod class. This may be more desirable than using the linearize class method, as the Linearizer object will allow more efficient recalculation (i.e. about varying operating points).""" if (self._fr is None) or (self._frstar is None): raise ValueError('Need to compute Fr, Fr* first.') # Get required equation components. The Kane's method class breaks # these into pieces. Need to reassemble f_c = self._f_h if self._f_nh and self._k_nh: f_v = self._f_nh + self._k_nh * Matrix(self._u) else: f_v = Matrix([]) if self._f_dnh and self._k_dnh: f_a = self._f_dnh + self._k_dnh * Matrix(self._udot) else: f_a = Matrix([]) # Dicts to sub to zero, for splitting up expressions u_zero = dict((i, 0) for i in self._u) ud_zero = dict((i, 0) for i in self._udot) qd_zero = dict((i, 0) for i in self._qdot) qd_u_zero = dict((i, 0) for i in self._qdot + self._u) # Break the kinematic differential eqs apart into f_0 and f_1 f_0 = self._f_k.subs(u_zero) + self._k_kqdot * Matrix(self._qdot) f_1 = self._f_k.subs(qd_zero) + self._k_ku * Matrix(self._u) # Break the dynamic differential eqs into f_2 and f_3 f_2 = _subs_keep_derivs(self._frstar, qd_u_zero) f_3 = self._frstar.subs(ud_zero) + self._fr f_4 = zeros(len(f_2), 1) # Get the required vector components q = self._q u = self._u if self._qdep: q_i = q[:-len(self._qdep)] else: q_i = q q_d = self._qdep if self._udep: u_i = u[:-len(self._udep)] else: u_i = u u_d = self._udep # Form dictionary of auxiliary speeds & and their derivatives, # setting each to 0. uaux = self._uaux uauxdot = [diff(i, dynamicsymbols._t) for i in uaux] uaux_zero = dict((i, 0) for i in uaux + uauxdot) # Checking for dynamic symbols outside the dynamic differential # equations; throws error if there is. insyms = set(q + self._qdot + u + self._udot + uaux + uauxdot) if any( _find_dynamicsymbols(i, insyms) for i in [ self._k_kqdot, self._k_ku, self._f_k, self._k_dnh, self._f_dnh, self._k_d ]): raise ValueError('Cannot have dynamicsymbols outside dynamic \ forcing vector.') # Find all other dynamic symbols, forming the forcing vector r. # Sort r to make it canonical. r = list(_find_dynamicsymbols(self._f_d.subs(uaux_zero), insyms)) r.sort(key=default_sort_key) # Check for any derivatives of variables in r that are also found in r. for i in r: if diff(i, dynamicsymbols._t) in r: raise ValueError('Cannot have derivatives of specified \ quantities when linearizing forcing terms.') return Linearizer(f_0, f_1, f_2, f_3, f_4, f_c, f_v, f_a, q, u, q_i, q_d, u_i, u_d, r)
def linearize(self, op_point=None, A_and_B=False, simplify=False): """Linearize the system about the operating point. Note that q_op, u_op, qd_op, ud_op must satisfy the equations of motion. These may be either symbolic or numeric. Parameters ---------- op_point : dict or iterable of dicts, optional Dictionary or iterable of dictionaries containing the operating point conditions. These will be substituted in to the linearized system before the linearization is complete. Leave blank if you want a completely symbolic form. Note that any reduction in symbols (whether substituted for numbers or expressions with a common parameter) will result in faster runtime. A_and_B : bool, optional If A_and_B=False (default), (M, A, B) is returned for forming [M]*[q, u]^T = [A]*[q_ind, u_ind]^T + [B]r. If A_and_B=True, (A, B) is returned for forming dx = [A]x + [B]r, where x = [q_ind, u_ind]^T. simplify : bool, optional Determines if returned values are simplified before return. For large expressions this may be time consuming. Default is False. Note that the process of solving with A_and_B=True is computationally intensive if there are many symbolic parameters. For this reason, it may be more desirable to use the default A_and_B=False, returning M, A, and B. More values may then be substituted in to these matrices later on. The state space form can then be found as A = P.T*M.LUsolve(A), B = P.T*M.LUsolve(B), where P = Linearizer.perm_mat. """ # Compose dict of operating conditions if isinstance(op_point, dict): op_point_dict = op_point elif isinstance(op_point, collections.Iterable): op_point_dict = {} for op in op_point: op_point_dict.update(op) else: op_point_dict = {} # Extract dimension variables l, m, n, o, s, k = self._dims # Rename terms to shorten expressions M_qq = self._M_qq M_uqc = self._M_uqc M_uqd = self._M_uqd M_uuc = self._M_uuc M_uud = self._M_uud M_uld = self._M_uld A_qq = self._A_qq A_uqc = self._A_uqc A_uqd = self._A_uqd A_qu = self._A_qu A_uuc = self._A_uuc A_uud = self._A_uud B_u = self._B_u C_0 = self._C_0 C_1 = self._C_1 C_2 = self._C_2 # Build up Mass Matrix # |M_qq 0_nxo 0_nxk| # M = |M_uqc M_uuc 0_mxk| # |M_uqd M_uud M_uld| if o != 0: col2 = Matrix([zeros(n, o), M_uuc, M_uud]) if k != 0: col3 = Matrix([zeros(n + m, k), M_uld]) if n != 0: col1 = Matrix([M_qq, M_uqc, M_uqd]) if o != 0 and k != 0: M = col1.row_join(col2).row_join(col3) elif o != 0: M = col1.row_join(col2) else: M = col1 elif k != 0: M = col2.row_join(col3) else: M = col2 M_eq = _subs_keep_derivs(M, op_point_dict) # Build up state coefficient matrix A # |(A_qq + A_qu*C_1)*C_0 A_qu*C_2| # A = |(A_uqc + A_uuc*C_1)*C_0 A_uuc*C_2| # |(A_uqd + A_uud*C_1)*C_0 A_uud*C_2| # Col 1 is only defined if n != 0 if n != 0: r1c1 = A_qq if o != 0: r1c1 += (A_qu * C_1) r1c1 = r1c1 * C_0 if m != 0: r2c1 = A_uqc if o != 0: r2c1 += (A_uuc * C_1) r2c1 = r2c1 * C_0 else: r2c1 = Matrix() if o - m + k != 0: r3c1 = A_uqd if o != 0: r3c1 += (A_uud * C_1) r3c1 = r3c1 * C_0 else: r3c1 = Matrix() col1 = Matrix([r1c1, r2c1, r3c1]) else: col1 = Matrix() # Col 2 is only defined if o != 0 if o != 0: if n != 0: r1c2 = A_qu * C_2 else: r1c2 = Matrix() if m != 0: r2c2 = A_uuc * C_2 else: r2c2 = Matrix() if o - m + k != 0: r3c2 = A_uud * C_2 else: r3c2 = Matrix() col2 = Matrix([r1c2, r2c2, r3c2]) else: col2 = Matrix() if col1: if col2: Amat = col1.row_join(col2) else: Amat = col1 else: Amat = col2 Amat_eq = _subs_keep_derivs(Amat, op_point_dict) # Build up the B matrix if there are forcing variables # |0_(n + m)xs| # B = |B_u | if s != 0 and o - m + k != 0: Bmat = zeros(n + m, s).col_join(B_u) Bmat_eq = _subs_keep_derivs(Bmat, op_point_dict) else: Bmat_eq = Matrix() # kwarg A_and_B indicates to return A, B for forming the equation # dx = [A]x + [B]r, where x = [q_indnd, u_indnd]^T, if A_and_B: A_cont = self.perm_mat.T * M_eq.LUsolve(Amat_eq) if Bmat_eq: B_cont = self.perm_mat.T * M_eq.LUsolve(Bmat_eq) else: # Bmat = Matrix([]), so no need to sub B_cont = Bmat_eq if simplify: A_cont.simplify() B_cont.simplify() return A_cont, B_cont # Otherwise return M, A, B for forming the equation # [M]dx = [A]x + [B]r, where x = [q, u]^T else: if simplify: M_eq.simplify() Amat_eq.simplify() Bmat_eq.simplify() return M_eq, Amat_eq, Bmat_eq
def to_linearizer(self): """Returns an instance of the Linearizer class, initiated from the data in the KanesMethod class. This may be more desirable than using the linearize class method, as the Linearizer object will allow more efficient recalculation (i.e. about varying operating points).""" if (self._fr is None) or (self._frstar is None): raise ValueError('Need to compute Fr, Fr* first.') # Get required equation components. The Kane's method class breaks # these into pieces. Need to reassemble f_c = self._f_h if self._f_nh and self._k_nh: f_v = self._f_nh + self._k_nh*Matrix(self._u) else: f_v = Matrix([]) if self._f_dnh and self._k_dnh: f_a = self._f_dnh + self._k_dnh*Matrix(self._udot) else: f_a = Matrix([]) # Dicts to sub to zero, for splitting up expressions u_zero = dict((i, 0) for i in self._u) ud_zero = dict((i, 0) for i in self._udot) qd_zero = dict((i, 0) for i in self._qdot) qd_u_zero = dict((i, 0) for i in self._qdot + self._u) # Break the kinematic differential eqs apart into f_0 and f_1 f_0 = self._f_k.subs(u_zero) + self._k_kqdot*Matrix(self._qdot) f_1 = self._f_k.subs(qd_zero) + self._k_ku*Matrix(self._u) # Break the dynamic differential eqs into f_2 and f_3 f_2 = _subs_keep_derivs(self._frstar, qd_u_zero) f_3 = self._frstar.subs(ud_zero) + self._fr f_4 = zeros(len(f_2), 1) # Get the required vector components q = self._q u = self._u if self._qdep: q_i = q[:-len(self._qdep)] else: q_i = q q_d = self._qdep if self._udep: u_i = u[:-len(self._udep)] else: u_i = u u_d = self._udep # Form dictionary of auxiliary speeds & and their derivatives, # setting each to 0. uaux = self._uaux uauxdot = [diff(i, dynamicsymbols._t) for i in uaux] uaux_zero = dict((i, 0) for i in uaux + uauxdot) # Checking for dynamic symbols outside the dynamic differential # equations; throws error if there is. insyms = set(q + self._qdot + u + self._udot + uaux + uauxdot) if any(_find_dynamicsymbols(i, insyms) for i in [self._k_kqdot, self._k_ku, self._f_k, self._k_dnh, self._f_dnh, self._k_d]): raise ValueError('Cannot have dynamicsymbols outside dynamic \ forcing vector.') # Find all other dynamic symbols, forming the forcing vector r. # Sort r to make it canonical. r = list(_find_dynamicsymbols(self._f_d.subs(uaux_zero), insyms)) r.sort(key=default_sort_key) # Check for any derivatives of variables in r that are also found in r. for i in r: if diff(i, dynamicsymbols._t) in r: raise ValueError('Cannot have derivatives of specified \ quantities when linearizing forcing terms.') return Linearizer(f_0, f_1, f_2, f_3, f_4, f_c, f_v, f_a, q, u, q_i, q_d, u_i, u_d, r)