BQ27542-G1: Gauge is going high while discharging (in remaining 10%)

The log files show that the gauge will adjust its capacity prediction close to end of discharge. The 2A use case re-adjusts predicted remaining capacity upwards, hence you get an upward change for unfiltered SOC (16% from 10%).

This indicates that the gauge underestimated capacity earlier in the discharge. This could be either due to incorrect cell resistance or load prediction.

What was the order of these tests? QMax is 3500mAh in these log files. This looks like the gauge didn't actually learn QMax (or it would be an extreme coincidence that it's 3500mAh, such a "nice" number).

Some adjustments of unfiltered values are expected but 10% to 16% looks too high for a constant current discharge. The gauge has a smoothing filter - filtered RM and SOC converge very well for all three discharges.

Please make sure that the gauge really completed a learning cycle. See www.ti.com/.../slua903.pdf

Dear Dominik,

Thank you for your support.

the customer did all learning cycle, they said that there is nothing special during leaning cycle. i requested log file, i will share it to you.

and I have more question,

1. is it normal that 680mA discharge and 3A discharge has no problem and only 2A discharge have problem? 2A value is between 680mA and 3A. i don't understand clearly.

2. you said about the whole capacity. is there any possibility that we did not use proper chemistry ID?

3. is the smooth filter already applied? or should I apply it manually?

4. does it have meaning that we do another learning cycle?

Best regards,

Chase

#1: If you check the log files, it's not just the 3A use case that shows the jump. It's just that the jump is more pronounced. Because gauging is a prediction, remaining capacity is subject to change by more than the previous prediction minus coulomb count for each prediction/simulation (which the gauge performs at various times during a discharge). Some predictions cause more, some less jump. This depends on load, load prediction accuracy, temperature, temperature prediction accuracy, cell model accuracy and many other factors.

#2: There's always the possibility that the ChemID isn't a good match. However, it looks more like the learning cycle wasn't done correctly. The gauge absolutely requires accurate QMax (3500mAh seems too "neat" for a learned value, though it's not impossible) because that will determine what OCV and cell resistance the gauge fetches from its internal chemistry tables. Also, cell resistance must be learned. At this point I suspect that the learning cycle wasn't done correctly.

#3: You can enable this with SmoothEn in Pack Configuration C. Then the gauge will report smoothed SOC.

#4: At the very least you should verify that one was completed successfully. I recommend keeping a log file of all registers during the cycle together with data memory log before and after to document the success of this learning cycle.

Dear Dominik,

Thank you for your reply.

Yes, I recommended doing 2nd learning cycle to my customer, very carefully with QMax value.

and I have a question for another learning cycle.

in the datasheet of Li-ion cell, nominal capacity is 3450mAh. can I input initial capacity to 3300mAh or 3200mAh when I start learning cycle?

or should I increase initial value?

and I attach the customer's log for first learning cycle. please review it if possible.

i900_learning cycle.xlsx

Best regards,

Chase