Forums » CircuitLab Support Forum

Hi. I'm simulating 160ms of transient response of a three phase powerline to an earth fault (i.e. a short circuit to ground). There are nearly 200 components (all passive - resistors, inductors, capacitors) in the powerline model (16 sections each representing 1km of line). I want to detect travelling wave effects so I need high frequency response.

Everything is fine if I use a time step of 10us or longer. However, once I try any step size below 10us, I get only the first 20ms of the run - when the current is zero. This is of no value to me. I need to see the whole 160ms for a step size of 1us or less.

I am using Firefox (latest version) on Windows 7 Home Premium 64bit (up to date). Hope someone can suggest something to help me.

Can you share your circuit? Also have you tried to run a simple circuit with the same three phase supply with the time step and duration needed? just some lunch time thoughts.

Here (I hope) is a link to a greatly simplified version with only 4km of powerline instead of the 16km I started with.

The problem remains. Time step of 10us is fine, time step of 1us produces no useful plot.

https://www.circuitlab.com/circuit/urs5j9/4km-tl

Yes this is interesting, when I run a simple RC circuit with the same time simulation parameters it works fine. The only thing I can think is that there is some numerical inaccuracy, what if to solve this you scale every thing up, like slow the frequency down by 10 and scale the components appropriately?

Hi Signality. Thanks for all that work. I have investigated the issue of negative capacitance and am not confident that it is the cause of the problem. I compared hand calculations and CL results for a simplified model containing negative mutual capacitances (see https://www.circuitlab.com/circuit/pxxw69/fully-lumped-100km-feeder-with-neg-mutual-caps/) and found that CL correctly computed the currents in the circuit. This is consistent with my understanding of numerical integration methods used by spice and other simprograms - they are essentially indifferent to the sign on the coefficients of the differential equations that they integrate. I have written a brief report on the comparison and will post it here if find out how to do so.

I suspect the increased meshing created by the mutual capacitance connections between phases may create a large increase in computational demand, though I have not seen huge solution times (some of my older complex models take much longer to solve at 1us time step, but do so and plot the output correctly). Increased computational demand should not in itself stop the plotting of output quantities I would have thought.

I'm at a bit of a loss to think of options other than there is some sort of bug that kicks in under certain conditions that are created by my model, and this stops the output being either plotted or exported in CSV form.

I would very much welcome any suggestions on what avenues I could follow up to get this resolved.

Sorry, that link doesn't seem to work. Try this one:

I think that this is really something that the guys at CL are best placed to comment on and try to fix.

Maybe the best option now is to raise this as a BUG REPORT (as described in: https://www.circuitlab.com/forums/support/topic/8s9n9hav/how-to-use-the-circuitlab-support-forum/)

I must apologise for muddying the discussion.

I should have re-read the notes on my own example circuit about negative inductance. The equivalence of a negative capacitance to an inductance is (of course!) only true at a single frequency and not over a band.

Seen over a range of frequencies, a Bode plot of your negative capacitances are simply capacitances with a phase response that is the negative of a normal positive capacitance.

As such all that stuff I said about replacing the caps with mutually coupled inductors would only apply at a single frequency and so would not really work well in a transient simulation where there is a relatively wide band of frequencies generated at the switching point.

Hi Signality. Yes, I'll raise a separate bug report. No apology required. In fact, due to your input I found and read the 1975 Laurence Nagel paper on SPICE - brilliant concept, beautifully executed. I now understand better how it all works.