Monday, January 17, 2011

A year with LiFePO4 batteries: What Have I Learned?

One year ago today I took my first drive in my Gizmo with 36 TS-LFP100AHA installed in a 2p18s configuration. See the blog posts in January and February 2010 for my setup. In that time I have learned quite a bit about LiFePO4 batteries. Playing with them, using them, measuring them, keeping an open mind about how they operate and avoiding the "conventional wisdom" about batteries are all good things to do. Being a physics type is also good but not necessary. It just means there is a chance that you learned that you record everything even if you don't think it is useful and especially if it doesn't support your current notions about things or something you are trying to prove.

In the past year the odometer went from 8640 miles and now sits at 14139 miles: one mile short of 5500 miles.
I also saw this nice pattern of numbers on the odometer! The aluminum over the speedometer was an attempt to shield sun from hitting the display. Sometimes I can't see how fast I'm going and don't want to get a ticket. I've since removed it since the CycleAnalyst has a speed function I can use for the infrequent times I need it.

I spent $46.80 on electricity to recharge and used 993.43kWh from the wall (representing 12,365.37Ah) to travel that distance. This comes out to a conservative average of 137Wh/mile from the battery pack. I'll post about current efficiency values in another post. I see a trend and think it might be tied to pack temperature so I'll post in the summer about it.

This is a CycleAnalyst from ebikes.ca showing the total cycle count, total Ah and total Miles driven. The miles is higher than my speedometer gives. I'm assuming that the CA is more accurate because I calculated it from 10 tire rotations but I wanted to keep a consistent count from before I installed the CA and after so I only record the speedometer values. Also, the difference in the CA total of 12,544Ah and my total of 12,365Ah is due to the fact that the CA records total Ah that the batteries deliver whereas my figure is from adding up all the individual trip Ah which include regenerative braking. From this you can calculate that regenerative braking gained me 1.4% range over this time period. [edit: I just realized that 7597Ah came from back calculating for the time I did not have the CA and are based on the kWh used to charge from the wall. A more accurate representative result would be a 3.7% gain from regen.] Not a whole lot, however I have gone over three times farther on this set of brake pads than I did before I had variable regenerative braking so it is definitely a benefit. Not to mention the ability to come down steep long hills and not have brake fade and the much quicker panic stops than without.

The Cycle Count of 290 represents the total number of times I have charged the battery pack until it was full and I reset the CA. The Ah value is not always exactly at zero after a charge. It ranges from a fraction of an Ah above or below to as much as 2.5Ah above or below depending on how long the drive was and how much the battery temperature changed. I think it is also dependent some what on how closely calibrated it is to the shunt. With my 500A shunt the CA only claims accuracy to 0.1A so at the end of charge when the Zivan ramps down to near zero amps it may or may not register on the CA. The 290 cycles represent an average (mean) discharge per cycle of ~21% with a median of 13.6%. The highest discharge amount was 89.9% (179.6Ah) and the smallest was 4.34% (8.68Ah). At this rate I hope the pack will last 10 years. Even then I will probably have spent more than if I had stuck with 6V flooded Golf Cart batteries. However, the utility of the vehicle is so much better that it is hard to put a price on it. Not to mention I am getting valuable information on a $5000 pack rather than on a $20,000 pack in a full size car. (I need lots of range for what I do with my car so I would need a much larger pack even though a $6500 pack would work well in a small car.)

One thing is for sure, when the range is more than 20 miles it is much easier to drive more. There were several times where the 20 mile range of lead acid were not enough for what I needed. I remember heading out to one of my rental houses to show it to someone and getting most of the way there and realizing I had forgot the key. I had to turn around and go home to get it. Unfortunately I didn't have enough range to make the trip a second time so I had to take an ICE vehicle. With LiFePO4 and an easy 60-70 mile range meant those situations are no longer an issue. I've driven from Kelso to Battle Ground, WA to work on a rental house knowing I would be there long enough to charge enough to make the return trip.

If you read my February 2010 blog and the two comments you will see there was talk about my original ending charge of 71V or about 3.94vpc. Jack Rickard correctly pointed out that charging this high would shorten the life of my cells. On July 28, 2010 at 11,365 miles on the odometer I installed my newly reprogrammed Zivan NG1 with the #612 charging algorithm for 19 LiFePO4 cells. I did adjust the voltage trim pot up so now the ending charge is 69.6-69.7V. At the same time I installed the remaining two pairs of cells. I top balanced all the cells using my BMS boards which shunt at 4.00V. This means my pack voltage was at 80V when I was finished. My DC-DC won't come on at this voltage and neither will my Sevcon PP745 controller. I used a bank of light bulbs I built as a load to discharge batteries and brought the voltage down to below 70V. This took very little time since there is very little energy above 3.45V with this particular chemistry.

One might be wondering about the top balancing if my BMS boards don't shunt until 4.00V. After all, 69.7V/20 cells = 3.485vpc. Well, since there is so little energy above 3.45vpc (with my 18 "cell" 200Ah pack it was only enough to get me up to about 15mph from a stop) there is really nothing lost by stopping earlier in the charging curve. The Zivan is incredibly consistent in stopping at its programmed cutoff voltage. It even will shut down if for some reason it can't control the voltage. I found out quite by accident on the first pack charge I did. Being naive about the characteristics of these batteries I went to the store thinking that since the voltage hadn't risen much it would still be a while before they were full. When I got back the charger was beeping an error code which turned out to be "unable to control voltage" or some thing like that. Zivan chargers may not be too great to end user adjustment but I sure am thankful that they just plain do what they are supposed to. They are truly plug-n-play if you don't make any changes to your battery pack. Another point about the small amount of energy at the top of the voltage curve. I've watched the voltage at the end of charge. From the time the charger first starts to ramp back on the current it is only 10-15 minutes before it is putting less than 500mA. This last phase (yellow light) of the charging curve lasts for 1 hour. Even if it continuously put in 500mA (it doesn't) that would only be 0.5Ah into a 200Ah pack. It comes on periodically and draws only 40W from the wall and that drops down rather quickly to under 20W and then shuts off. I'm sure if I didn't have any parasitic loads on the battery pack that it would shut off and not come on again during the remainder of the final charging curve.

Back to top balancing: I was wondering too what would happen. If these batteries drift I should start seeing it after a few charge cycles. I started recording the voltage at the end of charge when the charger was not charging. This was during the yellow light phase of charging. I could see the wall power on my kill-a-watt meter so I knew when the charger was not putting anything into the pack. When not running the charger draws about 4W. I can also hear when the charger shuts off and turns back on because I hear a high pitched squeak which is probably the PWM of the charger coming on or turning off. Other than the first time I took cell readings the charger was off. Below is a screen capture of the spreadsheet I'm using to record the information.

The first column lists the cell number and location in the pack. Cell pairs 19 and 20 were installed after the charger was reprogrammed. They only had about 3 cycles on them before being merged into the pack. I did have a mishap with pair #19 a day or two after installing them. I missed tightening one of the bolts which held the straps between this pair and pair #20. It had worked its way up about a centimeter or so. Amazingly, there was no melted metal but one cell of the pair took several Ah to bring it up to the rest. What I think was happening was that the high currents were draining the cell but the charge currents were not getting into it very well. It is possible that since I am using brass bolts rather than the stainless steel ones supplied by TS I avoided problems. I tightened this bolt down and charged this one pair with my bench top power supply. You will see that it is one of the higher voltage pairs, though not the highest, of the set. The next columns are a date followed by High/Low. the High/Low column shows an H for the highest cell and an L for the lowest cell for that day's reading. At the bottom of this column is the difference between these two values and represents the maximum voltage range of the pack. The cell immediately below the individual voltages is the sum of the voltages, below that is the total pack voltage as given my my ExTech Instruments EX830 multimeter. This is an awesome mid-range meter, BTW. Note that the ending voltage for the first reading is 70V. I lowered the ending voltage of my charger after this to keep my DC-DC happier. The next cell in the spreadsheet is the Watt reading on the Kill-a-Watt meter, below that is the pack temperature, in degrees Celsius, by an inexpensive indoor-outdoor digital thermometer. I placed the outdoor probe between the middle rows of batteries against the case. It is held in place by a folded up piece of closed cell foam to try to get a more accurate case temperature. The temperature was generally stable by the time the charging was finished so this value likely is very close to the internal temperature of the cells. The whole box is surrounded by 3/4" hard pink insulation foam as sold by home hardware stores. The box is by no means air tight but generally there is not much wind in my carport. Finally, the last cell is the average Volts per cell value.

As you can see, the voltages don't move around very much. The high and low cells don't change much. (Note that the dates are farther apart later in the table.) As Jack Rickard and others told me I'd get bored taking these measurements and eventually quit. Well, I like data and I'm out to test out a hypothesis that has been forming in my mind for a while. I'm no longer convinced I need a cell level BMS. I think a half pack voltage comparison is enough to spot a dying cell and take care of it. Before you tune me out read on.

After watching Jay Whitacare's talk to a group of Carnegie Mellon students and professors about their EV experiments I became less convinced that the risks of a cell level BMS were lower than a string level monitoring system. See the video linked to on the ChargeCar website or on the EVTV.ME blog. When I read "Shuttle reaction" in the comments to Jack Rickard's January 9, 2011 blog it finally hit me why these batteries are so different from other chemistries. I have since read about it in other places. It was hinted at in one of the responses to an EVDL post I made asking why my efficiency from the wall numbers were so much better with LiFePO4 batteries than with lead acid. A poster mentioned a secondary reaction in lead acid that didn't participate in the storage of energy but used energy in the reaction. I believe this was the shuttle reaction which provides the self discharge of so many other batteries. Lithium ion batteries in general and LiFePO4 in particular merely move a Lithium ion from one plate to the other during charge and back on discharge. There really isn't a chemical reaction, per say, taking place to make this happen. Barring a mechanical defect in the cell there is no way for self discharge to happen! When people put BMS boards on each cell, this then becomes the discharge mechanism for the cell, not the cell it self. If you have any LiFePO4 cells in storage make sure there is nothing on the terminals. If you want just check the voltage every 6 months or so. Unless there is something on the cell allowing charge to transfer between the terminals you will be pleasantly surprised at or bored with the fact that the voltage doesn't change much if at all, even out to the thousandth of a volt.

Another source for this information is on Gold Peak Industries (Taiwan), Ltd. website. In section 3.8 of the linked pdf it states,
“There is no shuttle-based self-discharge reaction in the Lithium Ion cell like that found in the NiMH and NiCd. As the cell ages, the self-discharge eventually becomes zero. Initially the cell suffers from irreversible capacity loss. This is a reaction of the electrolyte with the the active components if the cell. It occurs more rapidly with increasing temperature and cell voltage. For this reason, cells should not be stored fully charged at temperatures approaching 60°C. Optimally they should be stored at 25°C or less and between 30-50% state of charge. The lower limit is chosen because they are often stored in packs witch circuitry that demands a small drain on the battery. When one considers the circuitry needed for li-Ion, it becomes the most important source of self-discharge.”
Note that it states that the lower limit, 30%SOC, is chosen because so many put BMS circuits on them. Leave them off means almost zero self discharge. Furthermore, as the cell ages, what ever self-discharge there may be drops to zero any way. It sounds like the cells will become less and less prone to drift apart in any particular given pack.

Another sentence in the above quote caught my eye, "Initially the cell suffers from irreversible capacity loss." This may explain why my efficiency numbers were so high right after I installed my pack. Last winter wasn't as cold as this winter but it may be that as the capacity dropped in my cells it took a little more energy to put back what I used. Another, more likely, possibility is that each time I drove I didn't put back as much energy as I took out. I didn't have an amp hour counter at the time so I have no way to support either possibility. This may cause you to wonder how I came up with my total Ah numbers for my total. What I did was average the number of Ah/kWh out of the wall for the first dozen or so charge cycles after receiving the CA and then calculated the Ah used for all previous drives. The total I arrived at, number of cycles and miles driven were then entered into the CA to give reasonably accurate lifetime pack values.

So, what about my existing BMS boards? As you can tell I haven't been using the top balancing function of them. I think Black Sheep Technology did a very good job of designing a robust board so I don't expect any failures as some people have experienced with other products. These batteries do sag significantly when cold, so that a 1.5C discharge rate was enough to get the low voltage trip, LVT, to sound my low voltage alarm even though the pack was fully charged and had less than 1Ah removed. The total pack voltage was in the range of 58V so the 2.93 LVT value leads me to believe that maybe all boards were sending the alarm and not just a few. This isn't a big issue except that I'm about to install a switch to turn off the siren I installed so I'm not bothered by it when the pack is cold. If I forget it is off I may reverse a cell without realizing it. Or, I might get complacent and merely ignore the warning as a false positive and still kill a cell. I need to either remove the boards or at least the siren and install another type of monitoring system or leave the boards and install another type of monitoring system.

I'm planning on keeping the boards and installing another monitoring system. Why would I want to do this? Well, I could be asking for trouble but I don't expect my particular BMS boards to short out and drain a cell. They could quit working and not alert me to an over voltage or under voltage condition but I don't expect that either. The most likely failure would be the three wire connecting jumper wires running from board to board. If this happens they would fail to send the appropriate signal to alert me. I'm planning on buying or building and installing a Lee Heart type Batt-Bridge circuit. The difference is that I will either install a meter with zero at the center and have the needle swing to one side or the other if the two haves of the pack don't have the same voltage or I'll have a bar of LEDs where the middle one is lit and the ones to one side or the other will light up to indicate pack imbalance. The benefit of this type of circuit is that I can see what is going on under different load conditions. It is possible that under little or no load that the pack is balanced but that under load a weak cell will drop more than the others. I can then look for this cell knowing it is in one half or the other of the pack. It would also be possible to hook up a circuit to this which would turn off the charger on a pack imbalance condition during charging in case a bad cell wasn't caught before hand.

By having both systems in place, the cell level BMS and the Batt-Bridge type circuit, I can see what system gives me the information I need first. If the Batt-Bridge doesn't warn me before the BMS then it may not be the best method. If, the BMS doesn't warn me before the Batt-Bridge then it may not be the best method. If they warn me at the same time then the Batt-Bridge is the winner because it has fewer connections to the pack, only three wires, and it cost significantly less than the BMS system. Furthermore, it has fewer points of failure. It won't drain a single cell in the case of a short and it won't heat up from shunting current and potentially lead to a fire. While I hope I don't need either one this will hopefully provide one more data point to either support or deny the "conventional wisdom."

I hope this helps someone who is trying to decide how to manage their pack of LiFePO4 batteries in their EV. Moving to Lithium is a huge step up from using lead acid. It has very different characteristics than lead acid and what you learned about lead acid or other chemistries in the past needs to be either forgotten in its entirety and relearned for LiFePO4 or you need to re-check all previous notions against data for this exciting battery technology. If someone tells you that you have to do such and such or use this or that just ask, "how do you know that? What data do you have to support that?" You will be glad you did.

[EDIT: 01-19-2011 Be sure to read the comments. Also, if you didn't go read Jack Rickard's January 9, 2011 blog be sure you do and read comment #85 posted January 19, 2011]

Sunday, January 16, 2011

Rear Wheel Bearing Modification

I have two Gizmos and both have the same issue with rear wheel bearings. I assume that any Gizmo with a belt drive would have the same issue. I believe that due to the tension required in the belt, the weight on the rear wheel, loads due to acceleration and regen, road shock, and the narrow R12 bearing outer surface the aluminum cannot handle the pressure. I believe that it actually cold flows slightly allowing the bearing to start to move and continually rounds out the hub. On both Gizmo hubs the inner seal was warn down to the metal support due to the bearing slopping around as the wheel turned. It also caused a lot of noise.

As a proof of concept I took a pop can and cut a strip of aluminum just long enough and wide enough to slip into the groove made by the bearing. This worked just fine for about 50 miles and showed me that the noise was caused by the bearing movement. (Ok, for you physics types like me it was actually hub movement on the bearing but relative movement none the less.) Since this worked I decided to use a harder material so I bought some brass sheet and repeated the process. On the second try I finally got the piece made to fit and everything was fine for a while. Then one day I noticed that the outer seal was working out of the hub. Later it was even further out. Finally I took the wheel off and the seal just pulled out easily. What I found was that the brass sheet was working its way out from around the bearing and pushing out the seal. Time for a different approach.

Since the inner, belt side, wheel bearing also wears out quite fast I thought I would like to install a tapered roller bearing on each side and gain an adjustable bearing, more bearing surface, a wider outer race surface, and longer bearing life at the expense of a little more friction. The problem I ran into was that I couldn't find a bearing with the dimensions I needed. The aluminum hub would have to be machined out and the aluminum may not have enough strength to keep the outer race compressed. I confirmed this with a couple of knowledgeable bearing people. I then went to Waite Specialty Machine Inc. in Longview, WA and talked to Keith Warring. Keith is an awesome guy and knows his stuff. I took in one of the hubs to show him what the issue was. I explained what I had tried and what I had found about bearing options. He suggested that a steel sleeve be inserted and that two bearings be installed on the belt side. They regularly sleeve motors and other applications where the bearing seat has rounded out too large to properly hold the bearing. For these types of applications they machine the hole just slightly smaller than the sleeve then they heat the outer piece and freeze the sleeve and insert them. When the two are at the same temperature they create an interference fit where the sleeve won't move. They also cut and faced the spacer that goes between the bearings and installed the bearings so the hub was ready to go.

Note the extra piece between the bearing outer race and the hub. The picture below shows a different angle where you can see the sleeve.


The outer bearing where the castle nut goes did not show any signs of wear. In fact, the inner wheel bearing on all three wheels seems to be the first one to go. I believe this could be due to the fact that the shaft flexes a fair amount so those bearings don't get the same road shock.

The cost of all of this was a little over $500 for two hubs. As with much machining the setup cost is quite high so doing two hubs at once was much cheaper than doing one at a time.

The other issue I've had with Gizmo bearings is that the inner seal seat, the collar on the shaft, had rusted and then wears out the seal rapidly. I was able to sand down the collar, the metal is quite soft, and install a speedy sleeve which is a stainless steel sleeve that goes over the seal race to gives a good surface for the seal to ride on. These are not cheap! One cost me over $40 but I'm sure it was less than replacing the axle or the entire rear swing arm. I'm actually suspicious that the collar that the seal rides on were damaged first by a bearing failure which would lead to premature seal failure and then dirt and water damage. Inspect those bearings and seals regularly.