Inside View

Identifiable Components

• Microchip PIC18LF26K83, Microcontroller.  There is a reset push button and a small 5-pin ICSP (in-circuit serial programming) connector associated with the uC.  Features:

Program Memory Size: 64 KB

RAM: 4096 bytes

Data EEPROM: 1024 bytes

ADC maximum resolution: 12 bits

ADC Channels: 24

• Microchip 24AA256 (SMD marking code 4LD) is a 256K-bit I2C serial EEPROM.  256K bits is 32K bytes.  This is a reasonable amount of storage for extreme events, averages, failures, etc.  

• Texas Instruments BQ7694003, Analog Front End (AFE) for BMS.  Features:

Pure digital interface

Internal ADC measures cell voltage, die temperature, and external thermistor

A separate, internal ADC measures pack current (coulomb counter)

Directly supports up to three thermistors

Overcurrent in Discharge (OCD) detection

Short Circuit in Discharge (SCD) detection

Overvoltage (OV) detection

Undervoltage (UV) detection

• Infineon IAUS300N08S5N012 (SMD marking code A08S5N12) is an N-channel MOSFET rated 80V, 300A with a 1.2 mΩ on-state resistance.  A total of seven are used in the BMS. 

• Onsemi FAN3122 (SMD marking code 3122T) is a single 9-amp high-speed, low-side gate driver.  Located physically near the four parallelled MOSFETs.

A Surprise!

The copper plate under the heat sink measures 58 × 37 × 1.05 mm.  The resistance across its width dimension is a mere 0.03 milliohms.  By comparison, the area of the circuit board that it parallels measures 2.4 milliohms.  I'm guessing the same physical size PCB may be used in different AMOPACK products (where additional MOSFETs may be used).  In this application the heat sink probably assures good physical contact between the copper plate and the PCB — it certainly would not be needed to dissipate heat.

There is a thermistor (PCB silkscreen marked AD1) for temperature measurement.  It measures about 14k ohms at room temperature.  This would only be needed if additional MOSFETs were present under the heat sink.

Charge Switch

Another surprise.  The charge switch operates on the high side.  It utilizes a circuit configuration known as an “ideal diode” in which a diode's normal forward voltage drop is replaced by a low-resistance MOSFET path. 

A pair of paralleled MOSFETs on the high side connect the charger to the battery pack.  These MOSFETs are the same as those used on the discharge side.  Each has a 1.2 mΩ on-state resistance.  Using two in parallel provides a total resistance of 600 μΩ.

But using an N-channel MOSFET on the high side introduces an additional complication in that its gate drive voltage must exceed that of the pack voltage.  This problem is solved by the LTC4357, which is called a “positive high voltage ideal diode controller.”  The LTC4357's claimed benefits are a reduction in power consumption, heat dissipation, voltage loss and PCB area.  With the addition of two conventional diodes, the circuit also protects against reverse-polarity connection of the charger.  AMOPACK chose an SDURB3040CT two-diode array, rated 400V, 15A (pulse) for this purpose. 

Precharge Resistors?

Two 200-ohm, 5W resistors (connected in parallel) can be seen near the charger circuit.  They connect to the positive side of the battery and are likely intended to limit the surge current into the motor controller's capacitors.  

However, in comparing the charging waveform from a working Dragonfly battery, I was unable to correlate this resistance using other known values and the equation for RC-charging.  So, it's possible that the battery I tested has different resistor values.  It also occurred to the that pre-charging may be pulse-width modulated as well, but I saw no evidence of this in the waveform.

Switched-Mode Power Supply (SMPS)

The SMPS provides power from the battery to run the BMS.  The first component in the circuit is a resettable fuse.  It is marked “X756F 0LKZ” which seems to correspond with a PolySwitch device rated for a 750 mA holding current and a 60 volt maximum.

Another large component at the front of the SMPS is an Onsemi FQD5P10.  This is a P-channel MOSFET rated, -100V, -3.6A, 1.05 ohm on-resistance.  I'm thinking it's used as a high-side switch to selectively pass battery voltage to the SMPS.  Since my testing has been done with a lab power supply and no battery cells, conditions have not been met to turn this MOSFET on.  Therefore, I jumpered its source to drain, which caused the remainder of the SMPS to operate normally.

The actual heart of the SMPS is an LM5008MM (SMD marking code SAYB).  This is a DC-DC switching regulator running at about 120 kHz.  It rated for a maximum input of 95 volts and bucks battery voltage down to about 12 volts and can provide an output current of up to 350 mA.  This output is used for gate drive voltage to the low-side switching MOSFETs as well as providing input power to the next stage which is a linear voltage regulator. 

The linear regulator (Microchip MCP1703) provides 3.3 V to power all the digital circuitry.  It can accept up to a 16V input and is rated for a 250 mA output.

At a nominal battery voltage of 50 V, I measured the SMPS's input current draw at 7.5 mA.  This equated to a power consumption of 0.375 W (9 Wh per day).  That number is surprisingly close to my very rough accounting of a 10 Wh per day loss given that the battery's self-discharge rate is about 0.1V per day.

Previously, I had speculated that the Dragonfly's high self-discharge could be due to the BMS constantly attempting to balance the cell voltages.  I no longer think this is the case.

The BMS simply consumes that 9 Wh per day until it puts itself to sleep.  Sleep occurs when specific predefined conditions exist (for example, sufficiently low battery voltage).  The only way to awaken the BMS is then to connect the charger.  Although this may be reasonable behavior for a battery that sees constant use, it's not ideal for a battery that's powering an occasionally-used motor vehicle.

Learning from Samsung's INR21700-40T Datasheet

I found it curious that AMOPACK rates the entire battery for 500+ charge cycles, whereas 18650batterystore.com describes the individual cells as being good for “300+ charge cycles.”

AMOPACK undoubtedly knows more about this than I do.  In an attempt to reconcile the difference, I came across Samsung's “Confidential Proprietary” datasheet for the INR21700-40T.  As with most topics, the deeper you dig, the more complicated it gets. (Note that the datasheet was published in December 2017, and so may not contain the very latest information.  It was also marked “Tenative”.)

Most importantly, there are two different methods to describe the cell's characteristics: Standard and Rated.  The Standard method implies a more conservative usage (which yields more desirable numbers) than the Rated method.

The Standard Charge current is 0.5C (2 amps), whereas the Rated Charge current is 1.5C (6 amps).

A quick calculation reveals the Dragonfly's cells are being charged at about a 0.39C (1.56 amp) rate.  Here are the particulars:

• The INR21700-40T is a 4000 mAh cell.  (4000 mAh equals 4.0 Ah)

• The Dragonfly's charger is rated at 12.5A.

Having eight 4 Ah cells connected in parallel means a 1C charge rate would require 32A.  But there's only 12.5A available.  Thus, 12.5 / 32 = 0.39. 

Similarly, the cell's Standard Discharge capacity is a more conservative measure than its Rated Discharge capacity.

The Standard Discharge capacity is ≥ 4000 mAh, and defined as:

“...the initial discharge capacity of the cell, which is measured with discharge current of 800 mA (0.2C) with 2.5V cut-off at 23 ℃ within 1 hour after the standard charge.”

Whereas the Rated Discharge capacity is ≥ 3900 mAh, and defined as:

“...discharge capacity of the cell, which is measured with discharge current of 10 A with 2.5V cut-off at 23 ℃ within 1 hour after the rated charge.”
 

The test for Cycle Life is a worst-case condition described as being:

With Rated charge (6A, 4.2V, 100 mA cutoff) and

Maximum Continuous Discharge (35A, 2.5V cut off)

the Capacity after 250 cycles is ≥ 2400 mAh (60% of the standard capacity at 25 ℃).

Some Takeaways

• Unsurprisingly, the lifespan of a cell (or battery) depends on the severity of the duty it must endure.   

• The major factor in determining lifespan is depth of discharge (DoD).  A lesser DoD yields a longer life.

• End-of-life is generally considered to be a reduction to 80% of new capacity.

• Charging current may inherently be limited by the cost to manufacture (and power needed to operate) a high C-rate charger.

• For further reading, see: https://www.electricmotiontech.com/home/ev-tech-101/battery-care-and-feeding

• SNx5HVD308xE is a low-power RS-485 transceiver. 

• XP power IV1205SA is an isolated DC-DC converter module.  It accepts a nominally 12V input and produces 5V at a maximum of 200 mA. 

• Texas Instruments ISO7521CDW is a low-power dual-channel digital isolator.  It uses two internal capacitors to isolate the inputs from the outputs.

• Fairchild H11A817 is a 4-pin optocoupler.  Most likely for control of the RS-485 chip's transmit/receive line. 

In testing with a fully-operational battery, no single ASCII character at any typical baud rate elicited a response.  The communications protocol probably needs a data  “packet” to do anything.  Although I have a lot of experience with RS-485 communications, it would be nearly impossible to guess the proper packet sizes and payloads.  Furthermore, the protocol likely implements an error detection checksum (for which there are many possible calculation algorithms).

Resistance of Internal Conductors

The main discharge cables (positive red, negative black) are unmarked but comprise seventy-two 0.39 mm conductors.  This yields a cross-section of 10 mm2 and is equivalent to a #8 AWG cable.  I measured the resistance of a 350 mm length of this cable (including one crimped lug) at 0.64 mΩ. 

Interconnections between the cell pack and BMS PCB are made using four layers of 0.15 mm nickel strip spot-welded to a brass plate.  The brass is 1 mm thick and measured 30 mm wide x ~160 mm long.  This brass section exhibited a measured resistance of about 0.38 mΩ.  Back-calculating a value for resistivity of the brass puts Rho at about 72E-9 Ω⋅m (72 nanoohm-meter).  This implies the brass has a high zinc content (probably more than 30%).  Note that a high-resistance material is beneficial for spot welding because the welding current creates Joule heating.

Nickel strip exhibits a similar Rho value of about 70E-9 Ω⋅m.  Because the nickel's total thickness is only 0.6 mm (4 × 0.15 mm), it will have a proportionally higher resistance than the brass.  Thus, all four nickel strips together have approximately 1.67 (1.0 mm / 0.6 mm) times higher resistance than the brass plate. 

Quality Control

The photo below shows good weld penetration on the negative interconnect.  It was very difficult to tear the nickel off the brass, and I chose not to do so.  However, weld penetration on the positive interconnect was poor.  It was easy to tear the nickel off the brass.  At first, I wondered if this was due to there being so much more thermal mass in the much larger positive electrode.  But it seems that should not be a problem because each spot-weld is made very quickly (probably on the order of 100 milliseconds).

Bonus Photos