From: Ryan K. <rk...@si...> - 2015-12-04 21:52:17
|
My hackish solution to the immediate problem is to integrate 3 times at dt/3 within the for loop: Ndt = 3#factor to divide dt by dt_sim = dt/Ndt for i in range(1,N): e[i] = r[i] - y[i-3]#<-- one step time delay on the measurement #v[i] = Gc*e[i]#<-- this is only doing P control esum += e[i] v[i] = Kp*e[i] + Ki*esum + Kd*(e[i]-e[i-1]) #PID control if include_sat: v[i] = mysat(v[i]) t0 = dt*(i-1) #integrate Ndt times to get to the next dt for q in range(Ndt): t1 = t0 + dt_sim to, yo, xo = control.forced_response(G_int, [t0,t1], [v[i],v[i]], X0=X0) X0 = xo[:,-1]#<-- save for next time through for loop t0 = t1#<-- for next pass x[i] = squeeze(X0) y[i] = yo[-1] This keep my students moving forward for now. -- Ryan Krauss, Ph.D. Associate Professor Mechanical Engineering Southern Illinois University Edwardsville On Fri, Dec 4, 2015 at 3:21 PM, Ryan Krauss <rk...@si...> wrote: > Apparently I don't know which email list is subscribed. Sorry if this > comes through more than once. > > I am teaching a classical controls class for mechanical engineering > undergraduates. Up till now, I have tried to gloss over state-space. My > students need to do an initial condition simulation for a system that > includes actuator saturation. I have them do this by integrating for one > time step at a time using control.forced_response inside a for loop. We > are essentially doing a continuous time approximation of ZOH with the input > held constant for each time step. As the time step gets too large, sp. > linalg.expm has to use a higher order pade approximation and eventually > throws this error: > > ValueError Traceback (most recent call last) > > > /Users/rkrauss/git/python-control/zumo_PID_simulation_modified_ss.py in <module>() > > 103 t0 = dt*(i-1) > > 104 t1 = dt*i > > --> 105 > to, yo, xo = control.forced_response(G_int, [t0,t1], [v[i],v[i]], X0=X0) > > 106 X0 = xo[:,-1]#<-- save for next time through for loop > > 107 x[i] = squeeze(X0) > > > /Users/rkrauss/git/python-control/control/timeresp.pyc in forced_response(sys, > T, U, X0, transpose) > > 374 [np.zeros((n_inputs, n_states + 2 * > n_inputs))]]) > > 375 print('M=' + str(M)) > > --> 376 expM = sp.linalg.expm(M) > > 377 Ad = expM[:n_states, :n_states] > > 378 Bd1 = expM[:n_states, n_states+n_inputs:] > > > /usr/local/lib/python2.7/site-packages/scipy/linalg/matfuncs.pyc in expm(A, > q) > > 258 # Input checking and conversion is provided by > sparse.linalg.expm(). > > 259 import scipy.sparse.linalg > > --> 260 return scipy.sparse.linalg.expm(A) > > 261 > > 262 > > > > /usr/local/lib/python2.7/site-packages/scipy/sparse/linalg/matfuncs.pyc in expm(A) > > 580 > > 581 """ > > --> 582 return _expm(A, use_exact_onenorm='auto') > > 583 > > 584 > > > /usr/local/lib/python2.7/site-packages/scipy/sparse/linalg/matfuncs.pyc in _expm(A, > use_exact_onenorm) > > 635 if structure == UPPER_TRIANGULAR: > > 636 # Invoke Code Fragment 2.1. > > --> 637 X = _fragment_2_1(X, h.A, s) > > 638 else: > > 639 # X = r_13(A)^(2^s) by repeated squaring. > > > /usr/local/lib/python2.7/site-packages/scipy/sparse/linalg/matfuncs.pyc in _fragment_2_1(X, > T, s) > > 753 exp_diag = np.exp(scale * diag_T) > > 754 for k in range(n): > > --> 755 X[k, k] = exp_diag[k] > > 756 > > 757 for i in range(s-1, -1, -1): > > > ValueError: setting an array element with a sequence. > > > This is probably ultimately a problem for the scipy people, but my > students' project is due in 6 days. Any suggestions to quickly get the > attached simulation code to work for 60Hz simulation, i.e. dt = 1.0/60? > > > Thanks, > > > Ryan > > -- > Ryan Krauss, Ph.D. > Associate Professor > Mechanical Engineering > Southern Illinois University Edwardsville > |