• State Resolved
  • Locked Locked
  • Replies 13 replies
  • Answers 4 answers
  • Subscribers 63 subscribers
  • Views 591 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

BQ34Z100-G1: updating calibration data after learning

user5014632
Intellectual 515 points
Part Number: BQ34Z100-G1

Hi,

I have been doing 10 golden images for different batteries we can use in our products. All batteries are LiFePO4 batteries from the same manufacturer, 5 of them are 4SNP (12.8V nominal voltage) and the 5 other are 8SNP (25.6V nominal voltage) otherwise only the capacity is changing (from 27Ah to 96Ah).

I thus have all the gg.csv files exported after the learning cycle that I used to generate the golden images files.

However I now realize that the calibration data I used for these learning cycles were not optimum and I would like to modify them to increase precision.

As you can guess performing again these 10 learning cycles is not an option

Is it possible to modify (slightly, let say +/-5%) the calibration values in the gg.csv file before exporting again the golden image files? What could be the drawbacks of this method?

Looking forward to your insights.

Best regards,

Jeremie

  • 0
    TI__Mastermind 34775 points

    Hi Jeremie,

    There's no harm in modifying the calibration after learning cycle +/-5%, since the gauge will continue learning in the field and can adjust the learned resistance with previous calibration value by up to 15% during one discharge cycle. 

  • 0 in reply to Damian Lewis
    Intellectual 515 points

    Hi Damian,

    Great! Thank you for your quick answer on that topic.

    Best regards,

    Jérémie

  • 0 in reply to Damian Lewis
    Intellectual 515 points

    Hi Damian,

    I am realizing all the possibilities my end application has to get data from the gauge, either from the golden image flash stream or from the gauge itself.

    I could configure my battery charger directly from these data instead of using separate parameters in my host microcontroller.

    However I would have to modify additional parameters in the gg.csv after learning, such as Configuration>Data>Cell Charge Voltages, Configuration>Data>Charge Currents, Configuration>Data>JEITA Temperatures...

    Basically could you indicate which parameters are safe to modify after learning and which should remain unchanged?

    Best regards,

    Jérémie

  • 0 in reply to user5014632
    TI__Mastermind 34775 points

    Hello Jeremie,

    You can always configure the gauge features to match your application after the learning cycle. The learning cycle is related to the battery cell properties. As long as you don't change the battery chemistry that was used for the learning cycle, you're ok configuring the gauging features and tweaking parameters after the learning cycle.

  • 0 in reply to Damian Lewis
    Intellectual 515 points

    Hi Damian,

    Thank you for your quick answer.

    More than the chem ID, I bet the Ra tables should not be tinkered with either, are you sure there is no other register that should be preserved? (e.g. end of charge taper conditions)

    I want to make sure I won't jeopardize my learning.

    Best regards,

    Jérémie

  • 0 in reply to user5014632
    TI__Mastermind 34775 points

    Hi Jeremie,

    That's correct. The Ra and Qmax is part of the chem ID and what are learned during the learning cycle.

  • 0 in reply to Damian Lewis
    Intellectual 515 points

    Hi Damian,

    Thank you for this clarification.

    I read that as part of the golden image generation process it was advisable to average the calibration values over several boards (This i why I started asking about calibration modification after learning).

    Following the same reasoning, is it safe (or even advisable as well) to average Ra tables and Qmax values over several learning cycles of different production grade batteries ?

    Looking forward on your insight.

    Best regards,

    Jérémie

  • 0 in reply to user5014632
    TI__Mastermind 34775 points

    Hi Jeremie,

    user5014632 said:
    Following the same reasoning, is it safe (or even advisable as well) to average Ra tables and Qmax values over several learning cycles of different production grade batteries ?

    No you don't have to average Ra or Qmax because it's learned and updated in the field. The recommendation is to run an optimization cycle once you have your prototype build, if you want to do due diligence.

  • 0 in reply to Damian Lewis
    Intellectual 515 points

    Hi Damian,

    Since these values are learned and updated in the field I am wondering whether the learning cycle is really necessary.

    Actually I do have access to the real capacity (Qmax) of my batteries, and I notice that across different number of series cell and capacities the Ra tables are pretty similar (see below)

    Would it be possible in that context (same battery manufacturer, same cell model) to "build" a golden image by setting the Ra table, Ra flag and Qmax in the gg.csv file without performing a learning cycle?

    Best regards,

    Jérémie

  • 0 in reply to user5014632
    TI__Mastermind 34775 points

    Jeremie,

    I would recommend doing a learning cycle one of the cell first then you can fan it out appropriately to other cell configurations. The reason for this even though learning is done in the field it's filtered/limited. The resistance isn't allowed to change more that 15% per discharge cycle whereas learning cycle is unfiltered and without the restrictions.

  • 0 in reply to Damian Lewis
    Intellectual 515 points

    Hi Damian,

    Indeed, I see that the design resistance varies depending on the capacity of my battery pack (i.e. number of parallel cells). Is the learning cycle still necessary if we guess this value from previous measurements? i.e. is there anything else that the learning cycle compute?

    Best regards,

    Jérémie

  • 0 in reply to user5014632
    TI__Mastermind 34775 points

    Jeremie,

    Typically as capacity increases resistance decreases. For the same cells as you increase the number of parallel cells the resistance decreases. If you know the resistance (Ra values) for one cell and Qmax from a learning cycle you can scale appropriately for different battery series and parallel cells configurations.  As mentioned before Resistance and Qmax are what's learnt from the learning cycle.

    Can you close out this post?

  • 0 in reply to Damian Lewis
    Intellectual 515 points

    Hi Damian,

    Thank you for this confirmation, I feel more confident in using your products.

    Best regards,

    Jérémie