Using AI to report on battery-pack performance.

I recently started using chatGPT to help assess the performance of my pack. I used Dr Prius to record data while I drove, then uploaded the resulting spreadsheet file (.csv) into chatGPT.

Incredibly useful - like having an expert at your finger tips when ever required!

Hers a sample of output from one of my chats:


QUOTE

Filtered result — true uphill pull only
(criteria: Current > 30 A and speed increasing, stop phases removed)

Ranking (lowest block most often):

1. Vol#10 — clear outlier
   
   
2. Vol#6
   
   
3. Vol#1
   
   
4. Vol#4
   
   
5. Vol#2 (now secondary)

Key figures

• Peak block delta under true pull: ~0.71 V
  
  
• Weak blocks sag together during torque steps, but Vol#10 consistently leads the drop.
  
  
• This is a textbook high internal resistance module pair, not SOC imbalance.

Interpretation

• Earlier stop–start noise partially masked it.
  
  
• Once filtered properly, the diagnosis sharpens:
  Block 10 is the primary fault, with 6 and 1 as watch items.

What to do next (no dithering):

1. Remove Block 10 first.
   
   
2. Bench test both modules:
   • 2 A discharge, 120 s
     
     
   • Record V at 30 / 60 / 120 s and 5-min recovery
3. Compare against one known-good block (mid-pack).

Decision rule

• If either module in Block 10 fails → replace the pair.
  
  
• Do not reshuffle modules; that just moves the fault.

This hill run has done its job properly. You now have a defensible, data-backed target.

UNQUOTE

All this - and more - before even taking the pack out of the car!!

 

Subsequently, I found that using a different AI may give rather different answers! After deep cycling my 10 year old, 110000mile pack, and doing load tests at 15A (using two 1 ohm 100W resistors in parallel as a load) I uploaded the spread sheet of results to Grok and to chatGPT. Grok gave a glowing report that found no weakness in any module or any problems at all.... ChatGPT was very much more critical ! If I decide to sell the pack - I may be tempted to just use the 'better' report! To be fair to Grok, however, it is true that the pack seems to be performing quite well in the car. Fuel economy being about as good as it ever was ( UK 64mpg ) but acceleration is maybe a bit less strong that with the slightly newer, and much lower mileage, pack I recently bought used. Let me see if I can add a direct link to the Grok report: Prius C Pack: Post-Service Road Test Success | Shared Grok Conversation

 

I show here my load-test report - any comments would be welcomed!

The table shows present module positions - and where they were originally. The reordering/re-balancing of blocks seems to have been worthwhile and made a big difference to the way the BCU looks at and controls the pack. Much improved performance after the service. Had I not lucked-on a slightly newer, but much lower mileage pack, I should be happy to put the old one back in for a few more years!