After completing our single cell testing, we have a lot of data to go through. Hopefully this page will help guide people in going through the data. If you have any questions, contact Micah Black. The tests chosen were guided by the Individual Cell Testing Evaluation page.
...
I would love to have a graph generated for every single data point for every single cell. The Visualization Status is for this - do we have a graph of every cell for this measurement, as well as a graph of the distribution of the results (plot the result from every cell on a graph).
Open Circuit Voltage
Priority | 1 |
|---|---|
File Name | cell#-OcvTestScenario.csv |
Results Status |
Completed |
Visualization Status |
|---|
Completed |
These cells have been sitting for almost a year since being manufactured before we tested them. All cells have what is called a self-discharge current, where the cell gradually discharges itself. To determine the self discharge current cells are usually left sitting and their voltage periodically monitored. Any cells that diverge in voltage quicker than other cells have a high self discharge current.
...
Due to some measurement error and inaccuracies in controlling super low currents, the current measured by the SMU is jumps around a bit. We need to remove the measurements that current or voltage was an outlier, and then create somewhat of an average of the rest of the measurement with more weight being given to the voltage measurements taken when the current was closer to 0 - there is probably a more technical term for this, but you get the point. The distribution of sample data is below. Graphs were made in MatPlotLib in Python.
...
DC Internal Resistance Pulses
Priority | 4 |
|---|---|
File Name | cell#-DcCurrentPulseScenario.csv |
Results Status | Not Completed |
Visualization Status | Not Completed |
The DC Internal Resistance Pulse test is very common as it is super easy to do (maybe not so accurately) with a multimeter and a power resistor. However, there are many complications with this test.
...
Also notice that there is still a small slope during the linear region. This is the effect of discharging a cell - the voltage drops. We want to find the slope of the line and trace is back to when the pulse of current was first applied to the cell. This will tell up us the exact internal resistance of the cell, and not include any effects of the discharge voltage drop.
...
Note on this graph (the third graph), the bottom axis should read pulse number, and not amplitude of current. These pulses were all done at the same current.
More guidance for the DC Internal Resistance
...
Pulses:
...
This measurement is similar to determining the resistivity of a sample of material, but applied to a battery cells. We applied a changing current from -3A all the way to 3A and measured the voltage of the cell. This gives a fairly linear response, but would maybe be better split up in to charge and discharge sections, or possibly a quadratic would fit the data better.
The slope of the line on a voltage vs current curve gives you the internal resistance of the cell. It is essentially a DC pulse measurement, but average over all amplitudes of pulses.
AC Internal Impedance (1kHz)
...
The AC Internal Impedance test at 1kHz is an industry standard test for measuring the internal resistance of cells, and is usually the value that manufacturers put an upper limit on. (LG says <40mOhm no the MJ1 datasheet I believe). The AC test produces different results than the DC methods as the AC method also takes into account the inductance and capacitance of the cell, and some of the other complexities of the equivalent circuit model. Essentially, the voltage never has time to fully settle out.
The test happens by applying a 1kHz sinusoidal current (or voltage) to the cells, and measuring the response. In our case, we applied a 100mA sinusoidal current profile to the cell, and measured the voltage response.
The internal impedance is calculated by dividing the amplitude of the voltage by the amplitude of the current. We used MatPlotLib and python to fit our data to sinusoidal profiles and extract the parameters, then got our results.
Capacity Ration
...
I think that this test is super cool and has the possibility to give some super interesting results.
The term capacity ration is the capacity per percent state of charge.
Basically the idea here is to take a small chunk of the charge-discharge curve, remove any transients from the start of the test, then find the capacity drained from the cell during this small chunk of time.
Then, if we make the assumption that all cells will have a relatively similar state of charge vs OCV graph (or we can take the internal resistance in to account as well), and then match the slope of the discharge curve to an area on the full discharge of a single cell (which can be approximated as linear in small regions).
We can compare the capacity obtained from the reference cell to the capacity obtained from the 10 second discharge during this region, and get a capacity estimate for the cell.
Weight (Possible Indication of Capacity)
...
Some of the papers linked on the Individual Cell Testing Evaluation page found a correlation between weight and capacity, while others did not. For us, this data is just another source of outlier identification.
The cells are all super close in weight per the manufacturer's spec. We are expecting a maximum change in weight of under 0.5g over all 1400 cells.
Getting a number for the weight requires a bit of data processing, and understanding of how the tests were conducted.
Sample data for the weight measurements is shown in the graph above. The data will start with either the old cell on the scale, or no cell on the scale. If the old cell was on the scale, it is removed, and then the new cell is put on the scale. There will be measurement errors, so zoom in on the data in MatPlotLib to see what you're dealing with, remove outliers, and average it.
All testing result data can be found here in the Single-Cell-Testing-Results zip file: https://drive.google.com/drive/u/2/folders/19MzeWnzwiVnc4nDDM7SIhcqjXieO3bOU
Warning - We have a lot of data - The zip file is 500MB, and extracted it is over 6GB.
...
The voltage-current graph will show the pulses, in sequence and alternating from positive to negative as in the graph on the left. The DC internal resistance values calculated for each pulse are shown on the right.
...
For each pulse, we want to linearly extrapolate the voltage (using the last half of the voltage measurements for each pulse - green line) back to the start of the current pulse (black line). The black dot indicates where the voltage would have been at the start of the current pulse is parasitic transients are ignored.
The internal resistance in calculated using the following equation:
Internal Resistance (Ohms) = (V pulse - V rest) / (I pulse - I rest)
Where:
Value | Meaning |
|---|---|
V pulse | The steady voltage change, extrapolated to the start of the pulse - where the green line intersects the black line at the start of the pulse |
V rest | The voltage of the cell before the pulse was applied. For our calculations, this will be the open circuit voltage of the cell measured from the OCV test |
I pulse | This is the set current for the pulse. This can be taken from one of the current values during the pulse. Again, we can ignore the uA changes that may be happening, but a better approach would be to take the average of all of the points. |
I rest | We will assume this to be 0A, as that is what we set the DC load to during the rest times. It may be somewhere in the uA range due to regulation errors, but we will ignore those. |
As can be seen, the internal resistance changes as more pulses are applied. This is an effect of the ions inside the cells moving around in and out of the separator, causing the resistance to decrease. As I understand it, this is purely a function of the number of pulses applied, and not a function of the amplitude of the current pulse.
...
When you plot the internal resistance for each amplitude of current pulse, then we get the graph above. As you can see from this graph and from the second graph from the first set of three, the internal resistance eventually settles out to a constant value. The blue lines show the internal resistance settling. The green line shows the constant value of internal resistance for both charge and discharge pulses. We want to have both of these values saved to the csv output.
DC Internal Resistance Slope
Priority | 3 |
|---|---|
File Name | cell#-DcLinearSweepScenario.csv |
Results Status | Completed |
Visualization Status | Completed |
This measurement is similar to determining the resistivity of a sample of material, but applied to a battery cells. We applied a changing current from -3A all the way to 3A and measured the voltage of the cell. This gives a fairly linear response, but would maybe be better split up in to charge and discharge sections, or possibly a quadratic would fit the data better.
The slope of the line on a voltage vs current curve gives you the internal resistance of the cell. It is essentially a DC pulse measurement, but average over all amplitudes of pulses.
...
AC Internal Impedance (1kHz)
Priority | 2 |
|---|---|
File Name | cell#-AcInternalScenario.csv |
Results Status | Completed |
Visualization Status | Not Completed |
The AC Internal Impedance test at 1kHz is an industry standard test for measuring the internal resistance of cells, and is usually the value that manufacturers put an upper limit on. (LG says <40mOhm no the MJ1 datasheet I believe). The AC test produces different results than the DC methods as the AC method also takes into account the inductance and capacitance of the cell, and some of the other complexities of the equivalent circuit model. Essentially, the voltage never has time to fully settle out.
The test happens by applying a 1kHz sinusoidal current (or voltage) to the cells, and measuring the response. In our case, we applied a 100mA sinusoidal current profile to the cell, and measured the voltage response.
The internal impedance is calculated by dividing the amplitude of the voltage by the amplitude of the current. We used MatPlotLib and python to fit our data to sinusoidal profiles and extract the parameters, then got our results.
...
Capacity Ration
Priority | 6 |
|---|---|
File Name | cell#-TenSecondsDischargeScenario.csv |
Results Status | Not Completed |
Visualization Status | Not Completed |
I think that this test is super cool and has the possibility to give some super interesting results.
The term capacity ration is the capacity per percent state of charge.
Basically the idea here is to take a small chunk of the charge-discharge curve, remove any transients from the start of the test, then find the capacity drained from the cell during this small chunk of time.
Then, if we make the assumption that all cells will have a relatively similar state of charge vs OCV graph (or we can take the internal resistance in to account as well), and then match the slope of the discharge curve to an area on the full discharge of a single cell (which can be approximated as linear in small regions).
We can compare the capacity obtained from the reference cell to the capacity obtained from the 10 second discharge during this region, and get a capacity estimate for the cell.
Weight (Possible Indication of Capacity)
Priority | 5 |
|---|---|
File Name | cell#-weight.csv |
Results Status | Completed |
Visualization Status | Completed |
Some of the papers linked on the Individual Cell Testing Evaluation page found a correlation between weight and capacity, while others did not. For us, this data is just another source of outlier identification.
The cells are all super close in weight per the manufacturer's spec. We are expecting a maximum change in weight of under 0.5g over all 1400 cells.
Getting a number for the weight requires a bit of data processing, and understanding of how the tests were conducted.
...
Sample data for the weight measurements is shown in the graph above. The data will start with either the old cell on the scale, or no cell on the scale. If the old cell was on the scale, it is removed, and then the new cell is put on the scale. There will be measurement errors, so zoom in on the data in MatPlotLib to see what you're dealing with, remove outliers, and average it.
Data:
All testing result data can be found here in the Single-Cell-Testing-Results zip file: https://drive.google.com/drive/u/2/folders/19MzeWnzwiVnc4nDDM7SIhcqjXieO3bOU
Warning - We have a lot of data - The zip file is 500MB, and extracted it is over 6GB.
| View file | ||
|---|---|---|
|
| View file | ||
|---|---|---|
|
| View file | ||
|---|---|---|
|
| View file | ||
|---|---|---|
|
| View file | ||
|---|---|---|
|
| View file | ||
|---|---|---|
|
| View file | ||
|---|---|---|
|
Data Processing Scripts:
Here is the python file used to generate the AC-IR values. It is a bit of a mess. It should be run from the same file as the data is in as of right now, but can definitely be improved.
| View file | ||
|---|---|---|
|
...
...
| View file | ||
|---|---|---|
|
...
The graphs used for testing were made using this script:
| View file | ||
|---|---|---|
|