In-situ 18650 heater cell detection


MASSIVE restock of EV Batteries at www.batteryhookup.com 5% coupon "POWERWALLS"

enki

New member
Joined
Nov 9, 2019
Messages
29

1 Introduction​

This is the documentation of some experiments I performed in order to ascertain whether it's possible to detect cells generating excessive heat (aka "heaters") in a diy 18650 battery pack built using the popular 5x4 20.2 mm pitch 18650 cell holders. All of the findings should be applicable to bigger cell arrays as well.

2 tl;dr​

By using an array of LM75 temperature sensors attached to one side of a 18650 cell pack abnormal heating of a single cell to a temperature of at least 50 deg Celcius results in a measureable (2 deg C) temperature anomaly which may allow for heater detection before a major pack failure has chance to occur.

3 Experimental setup​

The experimental setup consists of a fake "pack" built out of 20 no-name 18650 cells connected (spot-welded) to a short strip of the 5P continuous nickel fuse roll from batteryhookup. The cells were welded only on the positive side with a small heater attached to the negative side of a single cell in order to simulate a "heater" cell. The cells used were junk cells having 0V and no charge in them for safety. The following image illustrates the experimental setup:
experimental-setup.png

The heater assembly consists of a 10 Ohm / 5W resistor attached to the negative side of a single cell with thermally-conductive adhesive. A MOSFET is used to regulate the current flowing through the resistor with PWM. An K-type thermocouple is attached to the cell next to the heater resistor in order to for a closed feedback loop to stabilise the heater temperature. The temperature from the thermocouple is being read using a MAX6675 breakout board. The top PCB contains five LM75 I2C temperature sensors soldered in a very simple chain:
ZIF-20C-TEMP-heater-locations.png

The temperature sensors are labeled U1-U5 and their relative locations can be seen on the PCB silkscreen. The location of the fake heater cell is indicated with the "HEATER LOC 1" text. During the experimental runs the temperatures read from U1-U5 sensors are logged into a CSV file for analysis. The assembled experimental setup can be seen below:
thumb-experimental-setup-photo.jpg
thumb-experimental-setup-heater.jpg

4 Software​

The experiment is being controlled by an Arduino sketch which performs the following functions:
  • regulates the heater temperature using a very primitive PID regulator driving a PWM output regulating current flow through the heater resistor
  • periodically reads out all temperature readings and prints them on the serial port
The code for the Arduino project can be downloaded for reference. The most critical part (the main loop) is shown below:
void loop() {
float temp = thermocouple.readCelsius();

float deltaT = target_temp - temp;

float pwm_drive = min( max(deltaT * P, 0), 1023);

// read temperatures for the ZIF-20C-TEMP board
float temp_U1 = U1.readTemperatureC();
float temp_U2 = U2.readTemperatureC();
float temp_U3 = U3.readTemperatureC();
float temp_U4 = U4.readTemperatureC();
float temp_U5 = U5.readTemperatureC();

Serial.print("target_temp[degC] = ");
Serial.print(target_temp);
Serial.print("\ttemp[degC] = ");
Serial.print(temp);
Serial.print("\tpwm[%] = ");
Serial.print(map(pwm_drive, 0, 1023, 0, 100));
Serial.print("\tU1[degC] = ");
Serial.print(temp_U1);
Serial.print("\tU2[degC] = ");
Serial.print(temp_U2);
Serial.print("\tU3[degC] = ");
Serial.print(temp_U3);
Serial.print("\tU4[degC] = ");
Serial.print(temp_U4);
Serial.print("\tU5[degC] = ");
Serial.print(temp_U5);
Serial.println();

analogWrite(heaterON, pwm_drive);
// For the MAX6675 to update, you must delay AT LEAST 250ms between reads!
delay(300);
}

The data printed by the Arduino code was monitored, plotted and logged using QtSerialMonitor which I unfortunately had to build from source code myself. An example data point gathered in the CSV file is presented below:
"time","target_temp[degC]","temp[degC]","pwm[%]","U1[degC]","U2[degC]","U3[degC]","U4[degC]","U5[degC]",
2022-01-09T15:01:18.890Z,70.000000,24.750000,100.000000,22.500000,23.000000,23.000000,23.000000,22.500000,

The columns convey the following information:
Column nameValue descriptionUnitNotes
timetimestampISO-8601 UTC
targettemp[degC]the heater setpointCelcius
temp[degC]the heater thermocouple temperatureCelcius
pwm[%]current PWM output%Rescaled from 0-1023 to 0-100%
U1[degC]current temperature for the U1 sensorCelcius
U2[degC]current temperature for the U2 sensor
U3[degC]current temperature for the U3 sensor
U4[degC]current temperature for the U4 sensor
U5[degC]current temperature for the U5 sensor

5 Experimental runs​

There were 3 experimental runs performed with different temperature setpoints for the heater but otherwise identical circumstances. Ambient temperature was not controlled but can be seen in the initial readouts from U1-U5 before the heater power supply was switched on. Each experimental run consists of four phases:
  • baselining - the temperatures are logged but the heater power supply is not yet switched on, used for logging baseline information from sensors
  • warmup - the heater power supply is switched on and the heater temperature starts to rise and stabilises around the target setpoint
  • observation - the data from the temperature sensors is logged for a few hours
  • cooldown - the heater power supply is switched off and the cell begins to cool down, experiment is again left for a couple of hours for the temperatures to again stabilise
The phases are not explicitly marked in the data but can be noticed when observing the relationship between pwm[%] and temp[degC]. When the heater power supply is switched off (in the baselining and cooldown phases) the temperature does not rise even though the PWM output is 100%. The summary of experimental runs with raw data linked are provided below:
Experimental runHeater setpointRaw dataNotes
Run 17009.01.2022_15.47.18.318Z_Log.csv
Run 26010.01.2022_14.57.09.844Z_Log.csv
Run 35011.01.2022_10.23.06.399Z_Log.csv

6 Data analysis​

The goal of the experiment is to ascertain whether an array of simple temperature sensors attached to a 18650 cell pack can successfully detect and/or locate a cell experiencing abnormal heat generation during charging or discharging.

6.1 Detecting a heater cell​

In order to move towards this goal first some analysis was done on the raw data to check if there is any difference between the temperatures detected by particular sensors in order to detect an anomaly. A Python script was developed for this purpose and ran against the gathered CSV data dumps:
import pandas as pd
import numpy as np

columns = [ "time","target_temp[degC]","temp[degC]","pwm[%]","U1[degC]","U2[degC]","U3[degC]","U4[degC]","U5[degC]" ]
data = pd.read_csv(csv_filename, usecols=columns)
data['min_temp'] = data[['U1[degC]', 'U2[degC]', 'U3[degC]', 'U4[degC]', 'U5[degC]']].min(axis=1)
data['max_temp'] = data[['U1[degC]', 'U2[degC]', 'U3[degC]', 'U4[degC]', 'U5[degC]']].max(axis=1)
data['temp_anomaly'] = abs(data['max_temp'] - data['min_temp'])

print(f'Data from {csv_filename}:')
print(data)
print('Temperature anomaly:')
print(data['temp_anomaly'].describe())

6.1.1 Experimental run 1​

Data from in-situ-18650-heater-detection/09.01.2022_15.47.18.318Z_Log.csv:
time target_temp[degC] ... max_temp temp_anomaly
0 2022-01-09T15:01:18.890Z 70.0 ... 23.0 0.5
1 2022-01-09T15:01:19.198Z 70.0 ... 23.0 0.5
2 2022-01-09T15:01:19.507Z 70.0 ... 23.0 0.5
3 2022-01-09T15:01:19.815Z 70.0 ... 23.0 0.5
4 2022-01-09T15:01:20.123Z 70.0 ... 23.0 0.5
... ... ... ... ... ...
69266 2022-01-09T21:01:52.570Z 70.0 ... 25.5 1.0
69267 2022-01-09T21:01:52.875Z 70.0 ... 25.5 1.0
69268 2022-01-09T21:01:53.193Z 70.0 ... 25.5 1.0
69269 2022-01-09T21:01:53.491Z 70.0 ... 25.5 1.0
69270 2022-01-09T21:01:53.797Z 70.0 ... 25.5 1.0

[69271 rows x 12 columns]
Temperature anomaly:
count 69271.000000
mean 2.648713
std 0.707099
min 0.000000
25% 2.500000
50% 3.000000
75% 3.000000
max 3.500000
Name: temp_anomaly, dtype: float64

6.1.2 Experimental run 2​

Data from in-situ-18650-heater-detection/10.01.2022_14.57.09.844Z_Log.csv:
time target_temp[degC] ... max_temp temp_anomaly
0 2022-01-10T14:01:10.676Z 60.0 ... 23.5 0.0
1 2022-01-10T14:01:10.982Z 60.0 ... 23.5 0.0
2 2022-01-10T14:01:11.289Z 60.0 ... 23.5 0.0
3 2022-01-10T14:01:11.595Z 60.0 ... 23.5 0.0
4 2022-01-10T14:01:11.903Z 60.0 ... 23.5 0.0
... ... ... ... ... ...
42364 2022-01-10T18:01:02.073Z 60.0 ... 26.5 0.5
42365 2022-01-10T18:01:02.379Z 60.0 ... 26.5 0.5
42366 2022-01-10T18:01:02.691Z 60.0 ... 26.5 0.5
42367 2022-01-10T18:01:02.996Z 60.0 ... 26.5 0.5
42368 2022-01-10T18:01:03.301Z 60.0 ... 26.5 0.5

[42369 rows x 12 columns]
Temperature anomaly:
count 42369.000000
mean 1.864240
std 0.648193
min 0.000000
25% 1.500000
50% 2.000000
75% 2.500000
max 2.500000
Name: temp_anomaly, dtype: float64

6.1.3 Experimental run 3​

Data from in-situ-18650-heater-detection/11.01.2022_10.23.06.399Z_Log.csv:
time target_temp[degC] ... max_temp temp_anomaly
0 2022-01-11T10:01:06.960Z 50.0 ... 25.0 0.0
1 2022-01-11T10:01:07.269Z 50.0 ... 25.0 0.0
2 2022-01-11T10:01:07.586Z 50.0 ... 25.5 0.5
3 2022-01-11T10:01:07.883Z 50.0 ... 25.0 0.0
4 2022-01-11T10:01:08.188Z 50.0 ... 25.0 0.0
... ... ... ... ... ...
58714 2022-01-11T15:01:39.845Z 50.0 ... 26.0 0.5
58715 2022-01-11T15:01:40.154Z 50.0 ... 26.0 0.5
58716 2022-01-11T15:01:40.462Z 50.0 ... 26.0 0.5
58717 2022-01-11T15:01:40.767Z 50.0 ... 26.0 0.5
58718 2022-01-11T15:01:41.075Z 50.0 ... 26.0 0.5

[58719 rows x 12 columns]
Temperature anomaly:
count 58719.000000
mean 1.545079
std 0.538658
min 0.000000
25% 1.500000
50% 1.500000
75% 2.000000
max 2.000000
Name: temp_anomaly, dtype: float64

As expected the observed maximum anomaly is larger when the heater setpoint is hotter:
RunHeater setpoint [degC]Maximum temperature anomalyNotes
Run 1703.5
Run 2602.5
Run 3502

6.2 Locating the heater cell​

The knowledge of the sensor board geometry and layout against the cells in the pack may allow for estimating the approximate location of a heater cell.
 

italianuser

Member
Joined
Feb 25, 2020
Messages
505
Oh this is a nice project. I don't know LM75, I use LM35. For a high numbers of cells could a "drop" temperature sensor, the ones you find in notebooks batteries, be more cost-effective (maybe with a lower accuracy in measurements)?
 

enki

New member
Joined
Nov 9, 2019
Messages
29
Oh this is a nice project. I don't know LM75, I use LM35. For a high numbers of cells could a "drop" temperature sensor, the ones you find in notebooks batteries, be more cost-effective (maybe with a lower accuracy in measurements)?
I know LM35 but it outputs a voltage which would require additional hardware for me to read (analog mux, adc, etc.). For an experimental setup using sensors which have a I2C bus was quicker to prototype. For a production device I will need to consider the BOM cost of having a bunch of simple sensors + analog mux and adc versus I2C or 1wire sensors. The tradeoff might be different based on the size of the pack.

I have a number of "drop" temperature sensors but they are simple thermocouples and they would likely require a dedicated ADC per sensor. With the voltages in those so small I don't think it's likely you can mux them without the offsets ruining your signal. So that gets much more expensive. Additionally it would be harder to assemble into a pack. With my "top PCB" solution you could even retrofit existing packs by just gluing or fastening the PCB to the nickel sheet.
 

enki

New member
Joined
Nov 9, 2019
Messages
29
Very nice looking work. What sparked (ha ha) your interest in this?
One of the threads on this forum (I lost the link) from hbpowerwall where he reported a pack suffering a major failure with one of the cells going thermal runaway. My hypothesis is that this doesn't happen suddenly but would be preceeded by abnormal heat generation from a single cell as the internal shorts develop. This makes it detectable. The safety features the community currenly implement did not detect or prevent this situation and discussions of monitoring individual cell temperature were followed by the report so I thought I'd give it a try.
 

italianuser

Member
Joined
Feb 25, 2020
Messages
505
I know LM35 but it outputs a voltage which would require additional hardware for me to read (analog mux, adc, etc.). For an experimental setup using sensors which have a I2C bus was quicker to prototype. For a production device I will need to consider the BOM cost of having a bunch of simple sensors + analog mux and adc versus I2C or 1wire sensors. The tradeoff might be different based on the size of the pack.

I have a number of "drop" temperature sensors but they are simple thermocouples and they would likely require a dedicated ADC per sensor. With the voltages in those so small I don't think it's likely you can mux them without the offsets ruining your signal. So that gets much more expensive. Additionally it would be harder to assemble into a pack. With my "top PCB" solution you could even retrofit existing packs by just gluing or fastening the PCB to the nickel sheet.
ops, sorry, I meant Dallas DS18B20 (suggested to me by @Wolf). Very smart TO-92, I imagine you do know it; I cascaded a series of them and read temperatures on a single Arduino/ESP8266 pin, and that is a great feature.

ds18b20.jpg

> I have a number of "drop" temperature sensors but they are simple thermocouples

Ok, I never tested them out, I do keep them aside when I open a notebook battery and checked the cost for buying 100, very cheap. For now I'm sticking to DS18B20, and will have a look at LM75
.
 

Redpacket

Active member
Joined
Feb 28, 2018
Messages
1,415
+1 for the one-wire DS18B20 devices.
Readings might be better with the sensor half way down the side of the cells ie between 4 cells but one sensor could likely sense in the middle of a block of eg 16 cells & give useful data on trending heat build up.
 

enki

New member
Joined
Nov 9, 2019
Messages
29
+1 for the one-wire DS18B20 devices.
Readings might be better with the sensor half way down the side of the cells ie between 4 cells but one sensor could likely sense in the middle of a block of eg 16 cells & give useful data on trending heat build up.
That could also work if there would be a way to reliably provide thermal contact between the sensor and 4 adjacent cells. There is a significant amount of space there, maybe you could wrap the sensor in something like some kind of heat-conducting compressible foam or smth :).
 

Redpacket

Active member
Joined
Feb 28, 2018
Messages
1,415
Maybe using the same adhesive pad stuff (eg on the flat side of the DS18B20 TO-92 package), sticking to the side of one cell?
Could be placed using a timber strip? Would be easy during the pack build & yes, more fiddly later!

Another position might be at the bottom (-ve end) of the cell as you'd get more direct contact with the cell.
Just thinking - wouldn't the heat transfer out the +ve end via CID & nickel strip might be less than optimal? But as your results showed, still got good data!
For test heating, maybe strap multiple smaller resistors (eg several 0.5W size) to side of a cell?
 

enki

New member
Joined
Nov 9, 2019
Messages
29
Very cool project!!! I know you are prototyping it. How would it work for larger pack setups? 14s80p? Can't wait to see what you come up with. Very good stuff.
Yes it would. The PCB I designed fits a 4x5 cell holder so for 80p you can just attach 4 of those side-by-side to a single block. The part that is currently missing is a "master" microcontroller that can aggregate the temperature readings from 4 separate sensing boards.
 
Top