Given a very low accuracy current sensor in a 3 phase inverter (in my case used for motor control) what is the best way to tackle the issue and keep reasonable performance and robustness at the same time of my feedback loop? The manufacturer of the sensor provides a low accuracy. Specs for my current sensor are the following:

  • Sensitivity: 33mV/A
  • Accuracy over full scale range: 12.8%
  • Full scale range $\pm$ 40 A
  • Offset is 1.65 V ($V_{cc}/2$, where $V_{cc}$ is 3.3 V)

Getting to how much this corresponds in A (assuming the full scale is on the analog V side) is about 1A, so I assume $\pm 0.5$ A as error.

The manufacturer does not state whether this is normally distributed error or anything else. I fear it might be a systematic accuracy error.

This is extremely course. Solutions that came to mind to close the loop without losing stability and keeping some sort of reasonable performance:

  • real time estimation/reconstruction of currents, given I have 3 sensors, one for each current (this is probably not doable because the switching frequency required for the application is too high and there is no time for digital filtering or estimation)
  • robust control ($H_2$ or $H_{\infty}$). My doubt on this is about the error modeling. As far as I understand robust control deals mainly with modeling the plant's parameters uncertainty and disturbance rejection.

Any idea to overcome this issue?

  • 1
    $\begingroup$ Define "low accuracy". You say the sensor accuracy is "low", and you say that you need to hold the current to an "accuracy of 0.5A" (presumably you mean $\pm$0.25A). But you don't say what accuracy the sensor can deliver. Nor do you say if the sensor's inaccuracy is due to random noise, or is systematic error. The devil is in the details, here -- please edit your question with further detail. $\endgroup$
    – TimWescott
    Jul 1 '20 at 14:45
  • $\begingroup$ @TimWescott thanks. Edited providing more details. Unfortunately the manufacturer does not state whether the inaccuracy is a random noise, with any distribution. I am hoping is a random normal noise, but worst case scenario I might assume uniformly ditributed random noise over the inaccuracy range. I would need a much better accuracy. Maybe swinging in a range $\pm 0.1A$ $\endgroup$
    – Nick90
    Jul 2 '20 at 12:08
  • $\begingroup$ I know I keep asking for more, but -- more please? Could you put up a block diagram, or schematic? It's not clear what you're controlling (torque? position?), or why you can't just control on what your sensor puts out. It may be that you just can't get there from here, and you need to change your current sensors. $\endgroup$
    – TimWescott
    Jul 2 '20 at 14:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.