Year Of DC Fast Charging And Battery Degradation Estimate
After about 1 year of SparkEV ownership, I started to log DCFC data for about a year, which resulted in 104 data points, average of exactly 2 DCFC per week. Coincidence or fake, you be the judge. I mostly use ABB chargers, which show percent, energy in kWh, and time. I started with recording the date and time and ambient temperature, but I got lazy and those records are spotty (only for the first one). But above data have been recorded on almost every DCFC session. What do you do with so much data? Why you generate figures, of course. We are visual creatures, and we love curvy figures. This is why strip clubs are profitable; we go there to stare at curvy figures! As with strippers, we鈥檇 rather look at real data, not fake ones (or pretend fakes are real). One way to do that is to look at the data from a known source. In this case, energy in kWh and charge time in minutes are strictly from the charger, and independent of the car.
That鈥檚 what I plot first. All those years of strip club training to spot fakes paid off as I can spot bimodal distribution right away in this graph. You can see that there are two (or more) distinct curves taking shape. This sets the tone for this blog post: major judgment call is based on strip club training. Therefore, this blog post should be treated as such. All data and analysis are completely subjective, probably wrong, and should not be trusted! I will give the raw data and source code to analysis, so you can judge for yourself how valid these may be. I find some strippers very pretty, so you may find the data and analysis presented here just as pretty. To get a better understanding, I infer average charge power (energy per unit time, aka division) over samples to see if there鈥檚 been any variations over time. Indeed, you can see a sharp increase in average power at about sample 35. What happened?
I did not change my charging locations or driving pattern, so it鈥檚 probably not the car or me. But ABB chargers used to update the data every second, and they changed to update every 10 seconds some time in the past. Then I go back to plotting charger-only data of energy vs time for samples 1 to 34 and 35 to end. Plots are lot better, though samples 35 to end looks like there鈥檚 yet another pattern. It could be that some DCFC units are not up to snuff as others? But a bigger question is, which data is closer to real, samples 1 to 34 or 35 to end? Apparently, I need more strip club training to spot the fake! Or in this case, I turn to even more powerful tool: guessing. I have another piece of data to help determine which may be real, which is the efficiency reported by the car. Over the course of over 19K miles, SparkEV shows 5.3 mi/kWh.
16K miles shown, but it hasn't changed. You can clearly see that samples 1 to 34 are much higher than later samples. Here, we look at the average for each. For samples 1 to 34, average is about 5.75 mi/kWh (cyan line) while the rest are about 4.9 mi/kWh (magenta line). But remember, DCFC is not 100% efficient. From previous blog post about SparkEV being the most efficient car in world history, we estimate about 94% efficiency for DCFC. 5 mi/kWh, red line). It seems first set of data to sample 34 is fake. It鈥檚 a not a problem; with advances in science, Dr. SparkEV will simply perform a surgery to make it come into shape as the real data so that power vs samples look pretty. There鈥檚 no fake data here! As with plastic surgery, making the data look good is subjective. One might think that since we know the average of both sets of data (samples 1 to 34 and 35 to end), simple scaling based on averages may work. Indeed, that鈥檚 what I did initially, but that still looked off. Best evenness is obtained by much trial and error, and that is shown in graph above.
We will use this new 鈥渟urgeried鈥?data for further analysis. There are many things one can infer from the data. First is various raw data over charge time, and some linear fitting. From the raw data, you can see that greater than 20 minutes of charging time result in lower percent and energy than a straight line fit. The curve fitting is performed on data that had ending percent less than 85%, not on the entire data. We鈥檒l go over that in next section. To get a better picture of charge taper, I plot average power (energy in kWh divided by charge time in hours) vs % charged. When the battery is above 80%, I would expect the power to dip, because that鈥檚 what I see at the charger if I鈥檓 staring at it. As you can see, there aren鈥檛 too many charge sessions that ended below 80% (blue dots).
