alxhslm / HarmLAB

Implementation of the harmonic balance method in MATLAB

HarmLAB

This toolbox offers an implementation of the Generalised Harmonic Balance method in MATLAB, supporting up to 2 base frequencies. This library can find periodic and quasi-periodic solutions of non-linear ODEs expressed in the following general form:

$$M_x \ddot{x} + C_x \dot{x} + K_x x + f_0 + f_{nl}\left({x},\dot{{x}},\ddot{{x}},{u},\dot{{u}},\ddot{{u}},\omega\right)={f}_e (t) = {M}_u\ddot{{u}} + {C}_u\dot{{u}}+{K}_u{u}$$

with an output given by:

$$y = g(x, \dot{x}, \ddot{x}, u, \dot{u}, \ddot{u} )$$

This toolbox has been developed to solve problems in structural dynamics, so much of the notation and terminology used stems from this field. However, it could equally be applied to problems in a variety of other disciplines.

The input is assumed to be of the following multi-harmonic form:

$${u} = A \sum_{k} \Re\left({U}_{k} e^{k\omega t}\right)$$

where the variable $A$ controls the excitation level.

The output is assumed to contain content at the same frequencies, so can be expressed as:

$${x} = \sum_{k} \Re\left({X}_{k} e^{k\omega t}\right)$$

This toolbox supports both Standard HBM with a single base frequency $\omega$ or Generalised HBM with 2 base frequencies which are interpendent:

$$ \begin{bmatrix}\omega_1 & \omega_2\end{bmatrix} = \begin{bmatrix}\lambda_1 & \lambda_2\end{bmatrix} \Omega $$

Analytical jacbobians have been implemented throughout to speed up the execution.

Problem definition

problem structure

The problem structure defines the system being simulated. This must have the following fields for the linear parts of the system:

• The mass, stiffness and damping matrices K, C, M
• The constant term F0 (which defaults to zeros)
• The matrices for the excitation Ku, Cu, Mu

The number of DOF and inputs is inferred from the dimensions of the matrices.

The non-linear parts must be specified via the model callback which must have the form:

function varargout = test_model(part,States,hbm,problem)
switch part
    case 'nl'
        %f_nl
    case 'output'
        %y
    ...
end
end

where part can be one of:

• nl which means this function should return $f_{nl}$
• nl_x, nl_xdot, nl_xddot which means this function should return $\frac{\partial f_{nl}}{\partial x}$ etc
• nl_u, nl_udot, nl_uddot which means this function should return $\frac{\partial f_{nl}}{\partial u}$ etc
• output which means this function should return the output $y$

The excitation must be specified via the excite callback which must has the form:

function U = test_excite(hbm, problem, w0)
...
end

and returns the $U_k$.

You can also store other useful information in the problem structure which is needed by your model (eg model parameters).

problem.res structure

For resonance problems (using hbm_res or hbm_bb), it is necessary to configure the term to be maximised. This can be of the following form:

$$ \frac{|X_k|}{|F_{e,k}|} $$

The res structure should have the following fields to set the numerator and denominator:

• iHarm which sets harmonic to use the terms from
• output which can be either x (or its derivatives),fnl, or none (to have no output).
• input which can be either u (or its derivatives), fe or unity (to have no denominator).
• NInput which selects which DOF from x to use
• NOutput which selects which index from fe to use

HBM structure

The other key structure needed to solve problems is the hbm structure, which stores information about the harmonics and other options. This has the following fields:

• harm contains information about harmonics
• dependence contains information about the form of $f_{nl}$
• cont contains settings for the continuation algorithm
• options contains settings for the continuation algorithm

harm structure

This must have the following fields:

• rFreqRatio is the ratio of the base harmonics to the base frequency. ie $\lambda$. For HBM standard problems rFreqRatio should be set to 1.0. For GHBM problems this should be vector.
• NHarm is the number of harmonics to include for each base frequency. This should have the same length as rFreqRatio.
• Nfft is the number of FFT points to use for the AFT transformation. This should be set to a power of 2 and have the same length as rFreqRatio.
• iHarmPlot sets which harmonic to plot on the FRF when using hbm_frf or hbm_bb. Typically you want to see the first harmonic so this should be set to 1.

dependence structure

This should have the following fields:

• x, xdot, xddot if the non-linearity has a dependence on the state and its derivatives
• u, udot, uddot if the non-linearity has an explicit dependence on the state and its derivatives
• w if the non-linearity has an explicit frequency dependence

The value should be set to 1.0 or True where there is a dependence, and 0.0 or False otherwise

cont structure

This is used to configure settings of the continuation algorithm. This should have the following fields:

• method: this should be one of the following values:
  • predcorr for a predictor-corrector algorithm (default)
  • none to simply step in frequency for hbm_frf or amplitude for hbm_amp and hbm_bb If method is predcorr, you can set the following additional settings in the cont.predcorr structure:
• solver to choose which non-linear solver to use (fipopt to use IPOPT or fsolve to use the inbuilt solver)
• step0 which sets the initial step size
• min_step/max_step which sets the minimum and maximum step size
• num_iter_increase/ num_iter_reduce which sets the threshold for increasing/decreasing the step size depending on the number of iterations
• C and c which sets the ratio to increase/decrease stepsize after a successful/unsuccessful step

options structure

This sets other more general options and can have the following fields:

• bUseStandardHBM: force the solver to use the standard HBM when there is a single base frequency.
• bVerbose: toggle whether to suppress output to console
• bPlot: toggle whether to suppress plots

Usage

Once problem and hbm have been defined, you must call the setup code:

[hbm,problem] = setuphbm(hbm,problem);

You can then call one of the following functions to solve a problem using HBM.

One-off problems

The most simple case is to find a periodic solution for a given excitation. Two different functions are provided to solve such problems, depending on the asumptions made about the base frequency:

• hbm_solve : this assumes the base frequency $\Omega$ is fixed to a known value. This is the simplest and most common case. This can be called as follows:

  sol = hbm_solve(hbm,problem,w0,A);
  

  where sol contains information about the solution including the components of $x$, $f_{nl}$ and $u$ at each harmonic
• hbm_res : this assumes that the base frequency $\Omega$ can vary in order to find the maximum response at resonance. This is configured by the problem.res field:

  sol = hbm_res(hbm,problem,w0,A,X0);
  

If you do not have initial guess for X0, then this can be omitted or set to [].

Continuation problems

Each of the different types of one-off problem can then be solved over a range of frequencies. If the base frequency is assumed fixed, there are two potential continuation parameters:

• hbm_frf: The base frequency $\Omega$ is used as the the continuation parameter, keeping the amplitude $A$ fixed. This yields a non-linear frequency response.

  sol = hbm_frf(hbm,problem,A,w0,X0,wEnd,XEnd);
  

• hbm_amp: The base frequency $\Omega$ is kept fixed, and the amplitude $A$ is used as the the continuation parameter.

  sol = hbm_frf(hbm,problem,A,w0,X0,wEnd,XEnd);
  

If the base frequency is allowed to vary, and chosen to satisfy an objective, then there is only one potential continuation parameter:

• hbm_bb: The excitation amplitude $A$ is the continuation parameter, and the base frequency $\Omega$ is chosen to satisfy the objective. This yields the "backbone" curve.

  sol = hbm_bb(bb,problem,A0,w0,X0,Aend,wEnd,XEnd);
  

If you not have an intial guess for X0 or XEnd, then this can be set to [] for any of these functions.