Some state estimators/observers in electrical drives applications need (as inputs) the voltages impressed on the machine terminals. Knowing that these voltages are always modulated using high frequency switching inverters makes the measurements of such voltages quite challenging. On the other hand using hard-wired low pass filters causes finite phase shift that can affect the estimator/observer accuracy. In light of these facts, what are the possibilities for making phase voltages available for real-time state observers/estimators ?
I hope I understood the question correctly. Let me know if I did not.
The control algorithm of the power electronics of the variable speed drive may estimates the voltage phasors from the voltage measurement. Some very well and some less well.
The three voltage phasors and the frequency of the three phase voltage can accurately, reliably and fast be estimated from a normal voltage measurement with relatively modest hardware. There is a small Finnish company MX Eletrix that produces power quality monitoring kWh-meters and they use a three phase implementation of the method described in
Koponen, Pekka; Seesvuori, R.; Böstman, R.. 1996. Adding power quality monitoring to a smart kWh meter. Power Engineering Journal, vol. 10, 4, pp. 159 - 163.
http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=532845
The meters are used for permanent power quality monitoring in distribution networks and customer interfaces. They internally have "real time" estimates of the voltage phasors. There is no phase shift but there is some latency both in estimation and communication. The main source of latency is now the internal and external communication of the meter. So far 100 ms (or longer) reading interval and about 1 s estimation time constant are used. Due to sparse sampling high frequency noise is aliased, which to some extent limits the possibilities to make the estimation of the voltage phasors faster. Sparse sampling rate is chosen so that all harmonic noise is filtered out and does not affect the estimation of voltage phasors and frequency. But now power electronics seem to be increasingly designed to produce interharmonic noise instead of harmoonic noise, just because there is still a gap in the interharmonics EMC requirements. The interharmonics actually cause more EMC problems than the same amount of harmonics.
Nowadays it is easy to use a much higher sampling rate and other approaches that rather completely remove the limitation I mentioned above. So we can go practically as fast as it is theoretically possible when taking into account the properties of the process noise in the voltage and the measurement noise due to the measurement hardware and the accurcy requirement for the voltage phasors. Thus latencies of the order of about 100 ms are possible to reach. Going faster requires compromising something.
Methods capable to similar performance are in some power electroni interfaces, but relatively lousy methods are also still applied in commercial power electronics.
Now I work with other topics and the above information is rather old. But I suppose the theoretical limitations remain the same. Communicatition may still be the main source of latency or cost, if the measurement and estimation methods are good.
Thanks Mr. Koponen, appreciating your response. Yes you understand the question correctly. Until I can download your IEEE article I will not be sure if this method represent a suitable answer to my question or not.
Anyway, I will elaborate more on my question. During the experimental phase in a study applying Luenberger observer for sensorless control of a permanent magnet synchronous motor, it was needed to inform the stator voltages among other variables to the digital signal processor which runs the observer algorithm. As a first approach, we used 2 galvanically isolated voltage senors connected to 2 very fast A/D channels of the DSP board. In fact the bottle neck was not the A/D sampling speed, but the calculation cycle of the observer which we could hardly shorten to only 160 micro second (this is 6.25 kHz)
Therefore, the voltage waveform which contains extremely short pulses beside relatively long pulses at multiple voltage levels(please refer to the attached waveform) could be sampled and read inside the observer routine at 6.25 kHz. Unfortunately I couldn't get any useful output of using such approach.
Then I assumed that the PWM inverter itself is merely an honest slave, meaning that it just executes what we tell it to do. In other words, we need not to measure Va, Vb and Vc, it is sufficient to assume that whatever the current control loop decides as reference voltages will be what we get (in average sense) at the inverter AC output.
To the extent that the voltage drop on the power electronic devices is small and the dead time inserted to avoid shoot-through in a single inverter arm is also negligibly small, this assumption can be considered acceptable.
In fact, the Luenberger observer could converge and the sensorless operation of the drive succeeded based on this rough assumption. I was wondering whether somebody would have a better and more accurate approach?
Try to see if Typhoon HIL 602 with 1us time step could help you out. Check it out at www.typhoon-hil.ch.
Best Regards
Thanks Luca. At 11 kHz space vector PWM, I have pulses ranging in width from below 4.5 micro sec to above 85 micro sec. Do you think that a fast "Hardware in loop" like Typhoon HIL 602 can execute a complex observer code in less than 4.5 micro sec?!
I think so. Its sampling time is 20nS and the time step can be as low as 0,5 or 1uS (each state space variable result is given at 0.5-1uS).
Dear Luca, have you any presentations or demo (pre-sales) to get approval of my department staff approval on purchase?
- Badges
- Science topic
- Similar topics
- Correction