Calculation of dead time and time constant for non linear system

Hello ,

I am simulating a non linear system on simulink . I used the scope to show the output and saved the output signal with time to the workspace . Is there any function to calculate the dead time and time constant given the output with time ?

Answer 1

[surya]

Calculation of dead time is a bit debatable aspect, you might want to look at a data based approach, like system identification.. System delay calculation. Might not be very accurate.. but gives an approximate idea.

Answer 2

[Rajiv Singh]

Delays are not necessarily separable from system dynamics such as effect of poles and zeros on the response. However, in many cases, DELAYEST in System Identification Toolbox works fine to deliver delays as number of multiples of data sample time. See also http://www.mathworks.com/products/sysid/demos.html?file=/products/demos/shipping/ident/iddemo3.html

[Ashim]

you can easily perform the time constant analysis on the output, if you know the objective non-linear function

   options = optimoptions(...); % select the options that match your data
  objfcn = @(y,t) y(1) + y(2)*(1-exp(y(3)*t)) - yobs; % objective function for time constant where y2 is usually the differential. use Taylor series to construct such an equation
  y0 = [y10, y20, y30]; initial guesses
  lb = [10, 10, 1]...; %lower bound
  ub = [100, 100, 10]; % upper bound
  [haty, resnorm, res, output, exitflag] = lsqnonlin(objfcn, y0, lb, 
  ub, options); non-linear least squares fitting

Answer 3

[Ashim]

you can easily perform the time constant analysis on the output, if you know the objective non-linear function

   options = optimoptions(...); % select the options that match your data
  objfcn = @(y,t) y(1) + y(2)*(1-exp(y(3)*t)) - yobs; % objective function for time constant where y2 is usually the differential. use Taylor series to construct such an equation
  y0 = [y10, y20, y30]; initial guesses
  lb = [10, 10, 1]...; %lower bound
  ub = [100, 100, 10]; % upper bound
  [haty, resnorm, res, output, exitflag] = lsqnonlin(objfcn, y0, lb, 
  ub, options); non-linear least squares fitting