Model Decomposition

class sails.modal.MvarModalDecomposition[source]

Object which calculates the modal decomposition of a fitted model derived from the AbstractLinearModel class. This is typically created using the initialise() classmethod which computes the decomposition and returns a populated object instance.

compute_modal_parameters()[source]

Compute the order 1 or 2 autoregressive coefficients associated with each pole or conjugate pair.

Warning

This function is a work in progress and has not been fully validated.

Returns:Array containing the time-domain AR parameters for each mode
Return type:ndarray
compute_residues()[source]

Compute the residue matrix from the eigenvectors associated with each pole.

Returns:Array containing residue matrices for each pole.
Return type:ndarray
excitation(resid_cov)[source]

Compute the mode excitation as defined in [Neumaier2001]. This is a metric to quantify the dynamical importance of each mode in the decomposition.

Warning

This function is a work in progress and has not been fully validated.

Parameters:resid_cov (ndarray) – Residual covariance matrix for modelled system
Returns:Vector containing modal excitation values
Return type:ndarray
get_mode_inds(fmin=0, fmax=None, mag_thresh=0, resid_thresh=0, index_mode='inclusive')[source]

Helper function to identify mode indices based on given criteria.

Parameters:
  • fmin (float) – Minimum frequency of modes to include (Default value = 0)
  • fmax (float) – Maximum frequency of modes to include (Default value = None)
  • mag_thresh (float ( 0 > mag_thresh > 1 )) – Minimum pole magnitude to include (Default value = 0)
  • resid_thresh (float) – Minimum value of mode residue norm to include (Default value = 0)
  • index_mode ({'inclusive','exclusive'}) – Flag indicating whether mode selection limits should be inclusive or exclusive (Default value = ‘inclusive’)
Returns:

Array of integers indexing the included poles

Return type:

ndarray

classmethod initialise(model, sample_rate=None, normalise=False)[source]

Compute the modal pole-residue decomposition of the A matrix from a fitted MVAR model.

Currently only implemented for a single realisation (A.ndim == 3)

Parameters:
  • model (LinearModel) – Fitted model object. Must be derived from AbstractLinearModel.
  • sample_rate (float) – sample rate of system in Hz (Default value = None)
  • normalise (bool) – Whether to adjust the phase of the eigenvectors (Default value = False)
Returns:

Modal decomposition object

Return type:

type

modal_freqz()[source]

Compute filter characteristics for each pole

modal_timecourse(data)[source]

Compute the modal time-course from a dataset and a fitted model. This is the transformation of a set of [nsignals x time] data observations into modal coordinates of shape [nmodes x time].

Warning

This function is a work in progress and has not been fully validated.

Parameters:data (ndarray) – Input data to convert into modal co-ordinates
Returns:Data transformed into modal time-series
Return type:ndarray
modal_transfer_function(sample_rate, modes=None)[source]

Compute a transfer function matrix based on the excitation period for each mode

Parameters:
  • sample_rate (float) – sample rate of system in Hz
  • modes (list of int) – List of mode indices to evaluate over (optional) (Default value = None)
Returns:

Transfer function computed for each mode

Return type:

ndarray

mode_indices

Returns a list of tuples of mode indices where each tuple contains the indices of either a pole-pair or an unpaired pole

Returns:list of tuples containing indices into the modes
Return type:List
per_mode_transfer_function(sample_rate, freq_vect)[source]

Extracts the transfer function for each mode by summing the transfer function across pole-pairs where necessary

The transfer function is computed for each pole using transfer_function() before summing individual modes (real valued poles or complex-conjugate pairs).

Parameters:
  • sample_rate – sample rate of system in Hz
  • freq_vect – Vector of frequencies at which to evaluate function
Returns:

Array containing the transfer function for each individual mode

Return type:

ndarray

period_hz

This is an old and deprecated way of accessing peak_frequency

pole_plot(ax=None, normscale=50, plottype='unitcircle')[source]

Plot the poles of the current system.

Parameters:
  • ax (matplotlib axes handle) – Optional axis on which to plot (Default value = None)
  • normscale (float) – Scaling factor for points on plot (Default value = 50)
  • plottype ({'unitcircle','eigenspectrum'}) – Flag indicating whether to plot results on a unit-circle or an eigenspectrum (Default value = ‘unitcircle’)
Returns:

Reference to axes on which plot was drawn

Return type:

matplotlib axes handle

residue_norm

The matrix-norm of the residue matrix of each mode

transfer_function(sample_rate, freq_vect, modes=None, sum_modes=True)[source]

Compute the transfer function in pole-residue form, splitting the system into modes with separate transfer functions. The full system transfer function is then a linear sum of each modal transfer function.

When sum_modes is False, this function returns the transfer function for each individual pole (ie will return a transfer function for each pole in a conjugate pair). Please use per_mode_transfer_function to compute the transfer function for individual modes (ie one transfer function for each real pole or complex-conjugate pair)

Parameters:
  • sample_rate (float) – sample rate of system in Hz
  • freq_vect (ndarray) – Vector of frequencies at which to evaluate function
  • modes (list of ints) – List of mode indices to evaluate over (optional) (Default value = None)
  • sum_modes (bool) – Boolean indicating whether to sum modes or return all (Default value = True)
Returns:

Array containing the transfer function for each individual pole

Return type:

ndarray