Re: Representation and Simulation of Dynamic Systems

*To*: mathgroup at smc.vnet.net*Subject*: [mg57002] Re: [mg56928] Representation and Simulation of Dynamic Systems*From*: "Caffa Vittorio Dr." <Caffa at iabg.de>*Date*: Thu, 12 May 2005 02:32:54 -0400 (EDT)*Sender*: owner-wri-mathgroup at wolfram.com

Hi Chris, First of all the topic here is "linear systems with time-varying coefficients". The response of such systems to deterministic signals can be computed with standard methods (e.g. NDSolve). If the input is stochastic (normally white noise) in the general case (non-linear systems) Monte Carlo techniques are generally used. But, in the special case of linear systems with time-varying coefficients, one can also perform a covariance analysis or use the method adjoints. In chapter 25 of the book: William S. Levine, Ed. The Control Handbook CRC Press there is an article on linear systems with time-varying coefficients with literature. Any book about the Kalman filter explains what covariances are and how they propagate. Cheers, Vittorio >-----Original Message----- >From: Chris Chiasson [mailto:chris.chiasson at gmail.com] To: mathgroup at smc.vnet.net >Sent: Wednesday, May 11, 2005 2:18 PM >To: Caffa Vittorio Dr. >Cc: mathgroup at smc.vnet.net >Subject: [mg57002] Re: [mg56968] Re: [mg56928] Representation and Simulation of >Dynamic Systems > >Dr. Caffa Vittorio, >Do you have a good link (http://....) that describes covariance >simulation and/or the method of adjoints? >Regards, > >On 5/11/05, Caffa Vittorio Dr. <Caffa at iabg.de> wrote: >> Chris, >> >> I have used as a toy example a dynamic system which happens to be linear >> with constant coefficients. Actually, I wanted to focus the discussion >> on non-linear dynamic systems. >> >> (As a special case, linear systems, possibly with non-constant >> coefficient, could be also analyzed with an extension of the method I am >> suggesting. Given the system description, one could for instance compute >> symbolically the matrices A(t), B(t), C(t), D(t). Then using the method >> of adjoints or a covariance simulation one could compute the system for >> example the system response to a stochastic input.) >> >> Actually the method I am suggesting is a surrogate of an interface into >> which one could place a block diagram. If such an interface were >> available I would probably do all my simulation work with Mathematica >> instead of using that-simulation-link-addition-which-must-not-be-named >> on that-mathematics-laboratory-program-which-must-not-be-named. >> >> Cheers, Vittorio >> >> >> >-----Original Message----- >> >From: Chris Chiasson [mailto:chris.chiasson at gmail.com] To: mathgroup at smc.vnet.net >> >Sent: Tuesday, May 10, 2005 4:22 PM >> >To: Caffa Vittorio Dr. >> >Cc: mathgroup at smc.vnet.net >> >Subject: [mg57002] [mg56968] Re: [mg56928] Representation and Simulation of >Dynamic Systems >> > >> >It seems like you are solving systems with constant coefficients. >> > >> >The trickiest parts would be (a) designing an interface into which one >> >could place a block diagram and (b) having Mathematica interpret that >> >diagram. >> > >> >If you could do that, control system professional can probably do >> >whatever else you might need ... >> > >> >http://library.wolfram.com/examples/CSPExtending/ >> > >> >Does anyone know if there is a good bond graph or block diagram >> >"package" available for Mathematica? >> > >> >I would really like to see the equivalent of >> >that-simulation-link-addition-which-must-not-be-named on >> >that-mathematics-laboratory-program-which-must-not-be-named for >> >Mathematica. >> > >> >Regards, >> > >> >On 5/10/05, Caffa Vittorio Dr. <Caffa at iabg.de> wrote: >> >> The behavior of (time-continuous, non-linear) dynamic systems can be >> >> numerically investigated with NDSolve. One can first sketch a block >> >> diagram of the system and then convert it into equations. Here is a >> toy >> >> example after the conversion: >> >> >> >> pos'[t] = vel[t] >> >> vel'[t] = -k pos[t] + force[t] / m >> >> >> >> This works fine if the variables are all states, as in the example >> >> above. But often, in order to describe a given dynamic system you >> want >> >> or you have to introduce some auxiliary variables (i.e. variables >> which >> >> are not states). This is in fact the case if you want to describe a >> >> generic dynamic system. Here are the standard equations: >> >> >> >> x'[t] = f[x[t], u[t], t] (state equations) >> >> y[t] = g[x[t], u[t], t] (output equations) >> >> >> >> where: x = state vector, u = input vector, y = output vector, t = >> time. >> >> In this case the components of the output vector are the "auxiliary" >> >> variables. >> >> >> >> I'm considering here a scheme for representing dynamic systems >> (possibly >> >> using a block diagram as a starting point) which allows the usage of >> >> auxiliary variables. This representation can be transformed into >> >> equations for NDSolve automatically. After having solved the >> equations >> >> it is possible to inspect not only the state variables but also the >> >> auxiliary variables. >> >> >> >> Comments or alternative solutions to the problem I'm considering are >> >> welcome! >> >> >> >> Procedure >> >> >> >> o) Sketch the system on a piece of paper. Here is a toy example: >> >> >> >> ---------- [ -k ] --------- >> >> | | >> >> V | >> >> force[t] --> [ 1/m ] --> + --> [ 1/s ] ---> [ 1/s ] ---> pos[t] >> >> | | >> >> | --------------> vel[t] >> >> | >> >> ---------------------------> acc[t] >> >> >> >> Note: [ 1/s ] is an integrator block >> >> [ k ] is a gain block >> >> >> >> o) Convert the sketch into a system description: >> >> >> >> In[1]:= sys = {pos'[t] -> vel[t], >> >> vel'[t] -> acc[t], >> >> acc[t] -> -k pos[t] + force[t] / m}; >> >> >> >> Note: the arrow points to the source of the signal. >> >> >> >> o) Make a list of the state variables: >> >> >> >> In[2]:= states = {pos[t], vel[t]}; >> >> >> >> o) Form the differential equations (the following steps could be >> >> performed by a function): >> >> >> >> In[3]:= lhs = D[states, t] >> >> >> >> Out[3]= {pos'[t], vel'[t]} >> >> >> >> In[4]:= rhs = D[states, t] //. sys >> >> >> >> force[t] >> >> Out[4]= {vel[t], -------- - k pos[t]} >> >> m >> >> >> >> In[5]:= eqns = Join[Thread[lhs == rhs], {pos[0] == pos0, vel[0] == >> >> vel0}] >> >> >> >> force[t] >> >> Out[5]= {pos'[t] == vel[t], vel'[t] == -------- - k pos[t], pos[0] == >> >> pos0, >> >> m >> >> vel[0] == vel0} >> >> >> >> o) Specify the parameters: >> >> >> >> In[6]:= params = {m -> 10, k -> 2, pos0 -> 0, vel0 -> 0, force[t] -> >> >> Sin[t]}; >> >> >> >> o) Solve the differential equations: >> >> >> >> In[7]:= sol = First[NDSolve[eqns /. params, states, {t, 0, 10}]] >> >> >> >> Out[7]= {pos[t] -> InterpolatingFunction[{{0., 10.}}, <>][t], >> >> >> >> vel[t] -> InterpolatingFunction[{{0., 10.}}, <>][t]} >> >> >> >> o) Inspect the results (including auxiliary variables) >> >> >> >> In[8]:= Plot[pos[t] /. sol, {t, 0, 10}] >> >> >> >> Out[8]= -Graphics- >> >> >> >> In[9]:= Plot[acc[t] //. sys /. params /. sol, {t, 0, 10}] >> >> >> >> Out[9]= -Graphics- >> >> >> >> Cheers, Vittorio >> >> >> >> -------------------------------------------- >> >> Dr.-Ing. Vittorio G. Caffa >> >> IABG mbH >> >> Abt. VG 32 >> >> Einsteinstr. 20 >> >> 85521 Ottobrunn / Germany >> >> >> >> Tel. (089) 6088 2054 >> >> Fax: (089) 6088 3990 >> >> E-mail: caffa at iabg.de >> >> Website : www.iabg.de >> >> -------------------------------------------- >> >> >> >> >> > >> > >> >-- >> >Chris Chiasson >> >http://chrischiasson.com/ >> >1 (810) 265-3161 >> >> > > >-- >Chris Chiasson >http://chrischiasson.com/ >1 (810) 265-3161

**Follow-Ups**:**Iterated NDSolve. HELP.***From:*Chrisantha <ctf20@sussex.ac.uk>