Evaluation of the Keysight E36731A Power Supply, Electronic Load, and Battery Emulator

Table of contents

RoadTest: Enroll to Review the Keysight Battery Emulator and Profiler E36731A

Author: Instructorman

Creation date: 21 Dec 2023

Evaluation Type: Test Equipment

Did you receive all parts the manufacturer stated would be included in the package?: True

What other parts do you consider comparable to this product?: GW Instek PEL-3031E 0-150V Electronic Load 300W - maybe

What were the biggest problems encountered?: Firmware anomalies, unexpected behaviors, awkward user interface. Nothing serious.

Detailed Review:

Element 14 Road Test of the Keysight E36731A DC power Supply/Electronic Load and BenchVue Battery Emulator application

This road test of the Keysight E36731A is entirely based on my user experience with the instrument in a specific application.  This review is not an evaluation of every function of the E36731A. Other reviewers do an excellent job of detailed functional evaluations and I recommend checking out their reviews.  For example, element14 member @saadtiwana_int’s review can be found here.

This review will be of interest to those that want to get a sense of how the E36731A performs in a real-world application that takes the instrument to some of its operating limits.  Please bear with me as I take time to describe the test case used to evaluate the Keysight E36731A for this road test.

Description of the test application

My interest in applying for the Keysight E36371A road test was driven by my enduring interest in off-grid solar photovoltaic (PV) power systems.  I have economic, environmental, and technical interest in PV systems that reduce carbon emissions. Two years ago, my wife and I decided to install 7.8 kW of roof mounted PV on our home.  We added a level 2 electric vehicle (EV) charger in the garage and this year we bought a Plug-In Hybrid Electric Vehicle (PHEV).  In the new year we are having a cold climate air source heat pump installed to hopefully reduce our natural gas consumption by 40%.

However, the system that supplied a test case for the E36731A is a small PV battery storage system that supplies power to a garden shed.  I have, for many years, been actively learning about off-grid PV systems.  My experimental platform is a garden shed where I have designed, installed, and repeatedly upgraded a small PV off-grid system.  The most recent upgrade in the summer of 2023 included installation of higher output solar modules (2 x 85W upgraded to 2 x 175W) and replacement of an absorbed glass mat (AGM) lead acid battery with a 100 Ah Lithium Iron Phosphate (LiFePO₄) battery.  I also added new loads to the system.  My gas-powered lawn mower was donated to charity and replaced with a battery electric lawn mower.  All my yard and power tools are battery powered now, and they all need to have their batteries charged, ideally with free power from the Sun.  I also added two e-bike chargers and a low voltage landscape lighting transformer to run a few yard lights at night.

My interest, therefore, in the E36731A is as a tool to characterize the new LiFePO₄ battery charge and discharge subsystem.  I will provide a brief description of the PV system in its current state, then devote the remainder of this report to how I used the E36731A to characterize three system components.

The functional capabilities of the PV system are listed below:

1. Provide 120 VAC power for
   1. A Wi-Fi security camera,
   2. A 24 VAC 200 W landscape lighting transformer,
   3. E-bike and power tool battery chargers,
   4. Other unspecified 120 VAC loads as needed.
2. Supply 12 VDC power for
   1. Three interior LED lights,
   2. One exterior LED flood light,
   3. An irrigation pump,
   4. A 12 VDC to 5 VDC switch mode regulator that supplies power to
      1. An STM32 control and monitoring system,
      2. A One-wire weather station,

• WS2812 LED exterior lighting,

1. Other 5 VDC loads as needed.

3. Manage energy storage and delivery via
   1. Two series connected 175 W monocrystalline PV modules,
   2. A 30 A maximum power point (MPPT) charge controller,
   3. A 100 Ah 12 VDC LiFePO₄ battery,
   4. A 1000 W pure sine inverter,
   5. Fuses, breakers, lightning protection, and service disconnect switches.

A block diagram of system components and interconnections is provided below.  Most of the system has been installed and evaluated.  I am still developing the STM32 based control and monitoring hardware.  The three components removed from the system for test bench evaluation of the E36371A are highlighted below with green shading.  They are an MPPT charge controller, a 1000 W pure sine inverter, and the 100 Ah LiFePO₄ battery.

Figure 1 Solar PV system block diagram

Photographs showing the interior mounted components, including the battery, fusing, disconnects and charge controller, are below.

Figure 2 Solar PV system components

Figure 3 Lithium Iron Phosphate battery

A more detailed description of the PV system and the characterization tests I carried out to evaluate a previous road test instrument (the Aim-TTi QPX750SP) can be found in this Element 14 blog post

Why the Keysight E36731A is a desirable tool for characterizing this PV system.

As a standalone instrument, the Keysight E36731A has two useful benchtop functions.  First, it is a versatile and moderately powerful DC power supply capable of outputting up to 30 V and up to 20 A within a 200 W power limit.  So, for example, you can’t drive 20 A at 30 V into a load because that would dissipate 600 W of power. The second especially useful function built into the E36731A is a controllable electronic load. The load can dissipate up to 250 W, sinking up to 40 A of load current at voltages up to 60 V.  Again, you can’t sink 40 A from a 60 V source because that would dissipate 2.4 kW.  Both the supply and the load have extensive data logging capabilities and the ability to use profiles to customize supply or load behavior over time. Both instrument functions were evaluated for this Road Test.

However, to really open the throttle on this instrument I found I had to use the Keysight BenchVue Advanced Battery Test and Evaluation application.  With this application running on the PC and a stable USB connection to the E36731A (more on this later) I obtained useful insights into system behaviors.

What features of the E36731A did I test?

In my road test application, I proposed a variety of possible tests that could be performed using capabilities in the E36731A.  As with any proposal that is future looking, circumstances at time of execution impact final outcomes.  Here are the features I was able to evaluate.

1. Battery profiler
2. Solar PV emulator (an application of the LIST sequencer feature)
3. Battery emulator
4. Programmable current slew capability

System parameters that define and constrain test cases.

Because I used the E36731A to better understand battery behavior in a specific PV system application, the charge and discharge characteristics specific to the system combined with E36731A specs to define tests that I could run. 

Charge characteristics

To emulate chare behavior with the E36731A it was important to know the PV rated output.  Rated values define the upper limit of available charging energy.  Specifications for the PV modules installed on the shed are given in the table below.

Figure 4 System energy supply specifications

So, these specs show that I have a problem. Two series connected modules can produce 36.2 VDC at 9.67 A.  That is about 350 W of power. The E36371A has a 200 W output power limit. To bring charge tests within the output range of the E36731A I simulated a single 175 W panel.  There is an instrument on my bench that can generate 350 W of power.  I ran a few charge tests with it to simulate two series connected modules.  I’ve included those tests for comparison purposes later in the review.

The benefit of a programmable output based on a list file.

In my view, rated output alone does not represent real-world conditions.  The Sun doesn’t digitally transition from zero to full irradiance at sunrise, nor does it digitally transition from full to zero irradiance at sunset. To obtain more realistic performance tests it would be great to use the LIST mode feature in the E36731A to simulate a typical day of solar irradiance behavior.  Unfortunately, where I live, a “typical day” is difficult to define.  At 53° N latitude we see enormous swings in daylight hours during the year from a low around 7 ½ hours near winter solstice to a high around 17 hours at summer solstice. This wide swing is compounded by the effects of daily cloud cover which can be 0% all day long up to 100% all day long.  Fortunately, I have access to daily power output logs for our house mounted solar array going back to the day it was installed.  These logs have 15-minute resolution and, upon review, clearly show the folly of the concept of a “typical day”.  What I decided to do was find logs that were recorded on completely cloudless days.  Example production logs from cloudless summer and fall days are shown below.

Figure 5 Production on cloudless summer day                                                                                                                                          Figure 6 Production on cloudless fall day

The kinks in the summer log might be caused by changes in atmospheric transparency and the tail at sunset is likely caused by scattering of light, or reflection off adjacent houses during our long lazy summer evenings.  Both curves kind of look like half-sinusoids to my eye.  I generated a half-sinusoid in Excel to overlay on the system production data.  The result is below.

Figure 7 Fitting a 1/2 sine curve to production curve.

The blue curve in the chart above is actual recorded system production on November 8, 2023, a cloudless fall day.  The orange curve is a computed half-sinusoid that spans the duration of system production that was greater than 0 watts.  The grey curve is the computed sinusoid shifted 45 minutes earlier to get a best fit.  Close enough for rock ‘n roll, as I like to say. Looks like a list file filled with computed sinusoid values will make a good proxy for solar PV emulation tests. Later in the review I will let you know how that idea turned out.

One quick, geeky, science-y aside.  There was a solar eclipse that passed over a good part of North America on October 14, 2023. Where I live the solar obscuration reached a maximum of 53.6% at 10:28 am.  It is interesting to note that of the six people observing the partial eclipse with me through my properly filtered and completely safe solar telescope, none noticed an appreciable decrease in brightness, even though more than half the Sun was obscured. Shadows sure looked different, but no one noticed a dimming of sunlight.  On the other hand, the solar array clearly recorded a decrease in solar energy.  Luckily, the day of the eclipse was nearly cloudless.   Look at the production curve from the day of the eclipse below.

Figure 8 PV response to partial solar eclipse

This is yet another example of why I like to compliment and compare my human sensory feelings with external physical measurement instruments.  In this case seeing wasn’t really a good reference for believing.  Anyway, moving on.

Discharge characteristics

Discharge is complicated because it can occur at any time of day or night.  During the day, the PV modules may provide some or all the load current, while at night only the battery provides load current.  To keep things simple, my battery emulation simulated nighttime operation.  The E36371A supplied power to the 1000W inverter based on a discharge profile obtained through the BenchVue Advanced Battery Test and Emulation application.  Below is a list of loads that I currently would like to support with the PV off-grid system.

Figure 9 System load demands

So, with all that background information covered, let’s get to the actual tests.  First up is my evaluation of the E36731A as a battery profiler.

Evaluation of the E36731A as a battery discharge profiler

My first test, prior to obtaining the BenchVue Advanced Battery application license, was to simply use the built in data logging capabilities to record a stepped current drain on the fully charged LiFePO₄ battery.  Per recommendations on page 35 of the E36731A user’s guide, I made up a pair of AWG 6 cables to attach the E36731A to the battery and torqued the binding posts to the recommended 8 in-lb, and then twisted both the load and 4-wire sense wires.   Note that the torque driver is upside down in the photo below, so it looks like it is showing 50.8 in-lb of torque.  It is, in fact, reading 8.05 in-lb.

N.B.: I noticed that after leaving the AWG 6 wires connected to the E36731A for a day (or more), that the screw terminals had loosened below 8 in-lb.  This I believe is caused by compression of the stranded wire bundle by the force of the blunt pin that presses into the wire strands.  I tinned the exposed strands with a reasonable amount of solder, but the loosening still occurred.  I recommend checking the tightness of the front panel binding adapter regularly and retorquing the connections if necessary.

Figure 10 8 in-lb of torque on binding post   

I took a quick look at the spec written on the front panel which told me max load current is 40 A and the max voltage is 60 V.  I thought, great, I’ll step up the load current to a value well below the maximum (25 A) with a battery voltage of 13.4 V and everything will be fine.  Nope.  When I tried to use front panel controls to step beyond 20 A the front panel display became erratic with current values jumping wildly above and below 20 A.  This encouraged me to stop the test at once and investigate further.

Although the electronic load in the E36731A has a maximum sink current specification of 40 A, the more important, and constraining, specification is power dissipation.  Power in the load is limited to 250 W.  This limit prevented my test sequence from achieving a 25 A draw from the battery when terminal voltage was ≈ 13.2 VDC (13.2 VDC x 25 A = 330 W).  At full battery charge (≈13.2 VDC) maximum E36731A load current is around 18.1 A assuming a maximum power dissipation of 250 W.  Here is what happened when I tried stepping up to a 20 A draw when V_bat = 13.4 VDC.

Figure 11 Exceeding max power spec at 20 A

While the instrument allowed me to dial in a load current of 20 A at a battery voltage of 13.4 V, the power dissipation exceeded the specified limit of 250 W by 17.77 W.  I only ran this test for about 30 seconds.  My concern being that at some point a thermal overload fault would shut down the output to prevent internal damage.  I limited later battery profiling tests to no more than 10 A at 13.4 VDC battery open circuit voltage.

My first successful battery profiling test using the Keysight BenchVue Advanced Battery Test and Emulation application was a gentle 5 A constant current (CC) profile from 100% state of charge (SOC) down to 88% SOC.  I experienced two unsuccessful profiling test attempts at 10 A CC.  These failed attempts are discussed under the heading Operational Issues toward the end of this review.

A screen shot of the successful discharge profiling test is shown below.  It is busy graphic rich with numeric and graphic information.  I needed to carefully study the image first to understand the data and then to extract meaning.  Even then I was left with a few lingering questions.

Figure 12 The BenchVue application display after a discharge profile test

There are, in my perspective, three main chunks of information.  The green battery icon with associated numeric data, the Voltage and Current graph on the upper right, and the State of Charge graph across the bottom.  My takeaways from the three highlighted areas follow below.

Figure 13 Battery status details

This section gives a snapshot of the battery state.  It looks like I might have stopped the test during an Ri measurement because V_t > V_oc and I is positive instead of negative.  But why? Why would the current change direction during a R_i check?  These questions led me to take a closer look at the R_i measurement pulses.  The image below shows current flow as logged by the BenchVue Battery application during an Ri measurement event at 10 A CC discharge.  The R_i test parameters were set to 1 s rest and 20 ms current pulse.

Figure 14 Detail of current flow during Ri determination

The R_i determination event starts when the E36731A stops drawing current. It then rests the battery for 1 second followed by a 20 ms draw of -10 A. These two steps can be used to measure V_OC and V_LOAD which allow an Ohm’s Law calculation of R_i.  Why the event finishes the way it does is not clear to me. After the 20 ms pulse at -10 A current drops to 0 A briefly, then jumps to +10 A for 20 ms before resuming constant discharge at -10 A.  Why?  Perhaps the E36371A is trying to replace the energy it removed during the 20 ms sense pulse?

Continuing with the information in the Battery Status window, the remaining capacity is an estimate because it seems to be based on the value I entered for battery capacity during test set up.  That is, I entered 100 Ah under Capacity Rating.  The application then seems to assume the battery starts the test at full charge and subtracts Ah capacity during the duration of the test.  The internal resistance value is calculated periodically during the drain profile.  The last calculated R_i value of 4.2943 mΩ seems impressive but is also to be expected for a nearly new 100 Ah battery.  Anything much higher would be concerning.

Figure 15 Log of I and V during discharge profile test

The V and I graph section logs battery voltage and current over the duration of the profiling test.  I discovered that it is important to make proper use of the scaling tools on the bottom right.  In this image the first 14 minutes and 19 seconds are missing because Auto Scroll was active, so old data was scrolled off the left edge.  The data is preserved in the log file, but careful use of the scaling tools is important to present data of interest.  In the next test I scaled to show all the test data.

One interesting take away for me from this image is that battery terminal voltage slightly increases for the first 80 minutes or so of the test while the battery is being discharged.  Not by much at all but I wasn’t expecting that.

The last section of interest is the SOC and R_i log.  This took a moment to figure out.  First, the SOC (bluish trace), plotted on the X-axis, runs from right to left.  That is, the starting SOC is on the right, and the finishing SOC is on the left.  The green trace shows calculated R_i.  It appears to very erratically during this test, but upon close inspection of the R_i scale on the left y-axis, the actual range of variation is only about 900 µΩ (4 mΩ to 4.9 mΩ).  So, R_i essentially remained constant from 100 % SOC to 88.5 % SOC.

Figure 16 Log of Ri and SOC during discharge profile test

I ran one more discharge profiling test at 10 A CC with a cutoff condition of SOC dropping to 80% from a fully charged state.  The results of the test appear below.

Figure 17 10 Ah CC drain with 20 Ah depletion stop condition.

This profiling test went well.  The cut off condition was specified as 20 Ah of consumed capacity.  With 10 A constantly draining from a fully charged 100 Ah battery I would expect to reach 80 Ah of remaining capacity in 2 hours.  The profiling test stopped at 2 hours and 47 seconds, generally supporting my expectation. According to the Battery Status window, the test stopped at 20.0017 Ah of Depleted capacity.  In the Settings window the Cutoff Condition was specified as 20 Ah of Consumed capacity.  While depleted and consumed are synonyms, I wonder if consistent use of one term might improve clarity. 

So, I now have a discharge profile for the 100 Ah battery from full charge down to 80% charge.  This will be sufficient to complete the rest of the aims in this review. I do intend to obtain more extensive discharge profiles for future use, but at a 10 A discharge rate it would take 5 hours to get to 50 % charge.  I like to be physically present during high current lithium battery discharge/charge tests.  I clearly remember a serious fire event several years ago at the polytechnic where I used to work that shut down the entire chemistry wing for months.  I heard that the cause was apparently a lithium battery being charged while unattended.

Evaluation of the E36371A as a solar PV emulator

For this review I decided to charge the battery through the MPPT charge controller, rather than directly from the E36731A, even though the E36731A can manage battery charge.  My aim is to use the E36731A to help me characterize the solar PV system, rather than just the battery.  The battery will not be charged from a steady state supply in the field.  Charge will always be managed by an MPPT controller whose energy input is highly variable sunlight.  I do gather data logs of real-world system performance and that information is useful.  However, with a programmable DC supply it should be possible to simulate cloudless days of any duration, or cloudy days, and everything in between.

My first bench charging tests through the MPPT charge controller were driven by an Aim-TTI QPX750SP programmable power supply, primarily because it could simulate the energy produced by two series connected PV modules as found in the real system.  Secondarily, I wanted to compare the output profile set up and data logging capabilities of the E36731A and QPX750SP.  I will discuss the performance of the Aim-TTi supply as a PV simulator first, then the performance of the Keysight supply. 

Setting up a half sine sequence with the Aim-TTi Test Bridge desktop application was a breeze. A screen shot of the tools available for doing so is shown below.

Figure 18 Aim-TTi Test Bridge application used to set up a 1/2 sine profile.

It is a simple matter to set up a custom output sequence by selecting from the available five generic waveshapes, then adjusting parameters like period, duration, voltage, phase, and offset. The example shown in the screen capture illustrates setting up a 3600 s (1 h) ½ sine wave. I was able to also set up a 12 h half sine wave sequence with ease. It did take the application about 15 s to compute all the values, but the set up was successful.

For a quick test I ran a 30-minute cloudless day charge on the LiFePO₄ battery when it was partially discharged. The data log files from the Aim-TTi supply were used to generate the chart below.

Figure 19 30-minute Cloudless Day charge trial

The supply ran through the programmed half sine sequence over a period of 30 minutes in a simulation of the sun rising and falling on a cloudless day. Peak voltage of 36 V is typical maximum output from two series connected 175 W PV modules at +25°C.  PV current was limited in the supply to 9.67 A, again simulating typical real-world performance. This simulation shows that the MPPT controller doesn’t start battery charging until PV voltage reaches about 16 V and it stops MPPT charging when PV voltage drops below 16 V. This agrees with my real-world observations of system behavior.

This short 30-minute test did not allow the battery to fully charge. The controller remained in MPPT mode for the duration of the test. I ran another 30-minute ½ sine charge with the battery closer to full charge.  You can see in the chart below that the battery reached full charge early in the test and switched from MPPT to BOOST mode.

Figure 20 Second 30-minute 1/2 sine charge filled the battery.

Using a “digital sunrise” to replace 20 Ah of battery capacity.

To return the battery to 100% following the last E36731A discharge to 80% I again used the Aim-TTi supply, but with a digital sunrise.  That is, a rapid stepping of supply voltage to 36 V in about 5 seconds.  This worked well.  The battery charged quickly.  Under these conditions the charge controller drew 9.67 A from the supply and boosted that to around 24 A to charge the battery. During the charge, the 10 AWG feed wires felt slightly above ambient to my touch, but I prefer to compare human sensory feelings with external instrument readings, so I took the thermal image shot shown below.

Figure 21 Thermal image of charge wires with 24 A current flowing

The S1 spot measurement on a charge wire shows a temperature of 27.2°C. Battery case temperature measured at 23.4°C. Ambient room temperature was 20°C. Although these temperatures and currents are well within operating limits, I still recommend that Lithium batteries never be left unattended during charging. Now, on to the E36731A PV simulation tests.

Setting up a ½ sine output profile on the E36731A

Spoiler alert, setting up a ½ sine output profile using the E36731A built-in LIST capability was, to be bluntly honest, frustrating and tedious.  After using the LIST sequence function, I have several suggestions for improving the user interface and functional capabilities.  These suggestions are detailed at the end of this review.

First, some background on the LIST function and how I planned to use it.

The Sequencer (List) function on the E36731A can support up to 512 data entries. Each data entry consists of a desired output voltage, output current, output step duration, and a choice to generate a digital signal at the Beginning of Step (BOST), the End of Step (EOST), or both. All these data elements are fantastic because they allow for wide flexibility in output waveshape, current limits, and profile duration.  It seemed that the E36731A would be able to reproduce the cloudless day simulation I performed with the Aim-TTi supply, with the caveat that the simulation would be for a single 175 W panel.  That isn’t a significant limitation in that single panel simulation supplies additional insight into system behavior.  I could not find any built-in tools for automatically generating waveshapes, so I set up an Excel spreadsheet to calculate sine values over 180° with a peak voltage of 18.1 VDC and a period of 1800 s (30 minutes) spread over 512 steps.  That was the easy part.  Getting 512 voltage values into the E36731A was not easy.  As far as I can figure out, there is no straightforward way to load an externally generated LIST sequence into the instrument.  The User Guide only describes how to manually enter values into the LIST sequence.  I have 512 voltage steps I’d like to load, along with current values, and step durations.  Is my only choice to enter all 512 voltages individually? Yes.

There was one shortcut.  The Add button copies values from the line above, so I only had to enter the current limit and step duration values once.  All I had to then was press Add 511 times, then, navigate back to the top of the LIST and individually enter 512 voltage values.  This, I managed to do, with only 6 mistakes, which were moderately easy to correct.  I will say, after using the front panel keys to enter well over 1500 keystrokes, the buttons have a nice tactile feel.  Touch is firm with a comfortable depth of travel, a confirming haptic click and a satin feel.  Top marks for button design. This is what the top few lines of the completed LIST sequence looks like.

Figure 22 A part of 512 LIST table entries for a 1/2 sine wave

At this point a mild terror overcame me when I thought “what if the E36731A doesn’t retain LIST values through a power cycle?”  I scanned through the User Guide looking for ways to save LIST values.  I found good news under User Menu – Store and Recall State.  Apparently LIST values are stored when the instrument state is saved.  Yay!

Then I found this intriguing detail: “Store Settings creates folders and files (.sta or .csv format) to store the instrument's state to the external memory, or to store the instrument's state to the internal memory.” (p.133 of Keysight E36731A Battery Emulator User’s Guide).  So, LIST values, along with lots of other instrument meta data, can be stored as .csv files?  That gave me hope that I could also load LIST values from a .csv file.  Alas, I could only get instrument settings to store in .sta format, which, when opened in Notepad was indecipherable to me.  Regardless, for the purposes of this review I was able to recover the full LIST sequence from a saved state file.

At this point I was ready to run the PV simulation.  I connected the E36731A outputs to the PV inputs on the MPPT charge controller, and the battery to the Battery terminal on the charge controller.  Out of an abundance of caution I enabled Over Voltage Protection (OVP) and Over Current Protection (OCP) on the E36731A simple because I had never used it in this mode before and I didn’t want any surprises.  Every time I ran the simulation OCP tripped as soon as output voltage rose through 15.8 VDC.  This is the point where the charge controller checks to see if it can enter MPPT mode to start charging the battery.  It does this apparently by trying to draw a lot of current from the PV module.  I adjusted the OCP trip delay incrementally up to the maximum amount (255 ms) and still the OCP activated at 15.8 V.  My solution was to turn off OCP.  That solved the problem.  The test finally ran without incident for the whole 30 minutes.  Result of the test in graphical form are shown below.

Figure 23 Results of a 30-minute Cloudless Day PV simulation on the E36731A

Here are two system behavior insights I obtained by using the E36731A in this test:

1. A single panel doesn’t generate enough power to fully activate the current boost capabilities of the charge controller. Once MPPT charging started at around 18.1 VDC, the charge current to the battery (measured by an external clamp meter) was almost equal to the current drawn from the E36731A.  Simulations of two series connected panels showed the charge controller boosting charge current far above panel current as voltage went from 18 V to 36 V.
2. The MPPT controller periodically checks PV panel operation all the time. See detail chart below which shows a part of simulated sunrise when PV voltage is around 2 V.  I did not see this detail when using the Aim-TTi power supply.

Figure 24 Detail of MPPT controller operation at low PV voltages

There are several improvements to the LIST sequence function that I have noted at the end of this review under Functional Improvement Recommendations.

Evaluation of the E36371A as a battery emulator

The tests in this section allowed me to use the E36731A as a simulated battery to power a 1000 W DC to AC inverter.  Clearly, the E36731A cannot provide enough power to fully simulate a 100 Ah battery, but it can provide enough power to run some of the smaller loads in the PV system.

First step, characterize inverter cold-start

The inverter I am testing is new.  I have never powered it up before, so my first tests characterized its start up behavior using the E36731A as a stand-alone datalogging power supply.  There were no AC loads on the inverter for these tests.  My earlier experience with DC to AC off-grid inverters is that they can draw a lot of current briefly at power up, especially if they have been shut down for some time.  I call this a cold-start condition.  With a 200 W source limit on the E36731A and V_out set to 13.2 V to simulate a well-charged battery, the steady state current limit is 15.15 A.  I expect I can temporarily draw more than that during start up.  Because I do not know how the inverter will behave during start-up, I activated OCP protection trips, starting at 10 A, just to get to know this new inverter.  OCP tripped on power up at 10 A, 12 A, 14 A, and 16 A.  It stopped tripping at a setting of 18 A.

I used the built in Scope/Data log feature on the E36731A to get a record of what happens during a cold start.  Here is what the data log revealed.

Figure 25 1000W Inverter cold start current draw

With 200 W available from the E36731A to drive the inverter I decided to load the inverter with the 60 W Power Tool battery charger charging a depleted 10 Ah battery pack.  A photo of the test set up is below.

Figure 26 E36731A battery emulation test set up

The E36731A DC output is connected via AWG 6 cables to the 1000 W inverter. A Wi-Fi smart power plug with power consumption monitoring is connected to an AC outlet on the inverter and a power tool battery pack charger with a depleted 10 Ah battery is plugged into the smart plug.

Rated power consumption of the battery pack charger is 60 W, but in this case, it drew only around 27 W. During this test I found the Keysight BenchVue Battery Test and Emulation application supplied useful insights into inverter performance.

A benefit of using the BenchVue application for discharge emulation is that once a battery profile has been created, the profile can be loaded and parameters like initial state of charge can be easily typed into the application.  You don’t have to wait through a battery recharge or discharge cycle to start a new test, just enter new values into the application and press Start. For this test I loaded a previously generated discharge profile obtained from the LiFePO₄ battery, then set initial state of charge to 95%. I checked load power consumption using the smart plug and monitored current draw into the inverter using a clamp meter. The test was sequenced as follows:

1. With no loads attached to the inverter and with the inverter powered off, start the emulation from BenchVue.
2. Turn the inverter on and let it stabilize.
3. Plug the battery pack charger into the smart plug and let it stabilize.
4. Insert the depleted battery into the charger and gather charge data for one hour.
5. Remove the battery pack.
6. Unplug the battery pack charger.
7. Shut down the Battery Emulation test from within BenchVue.

The State of Charge graph in the application tracks loaded State of Charge and Ri. Sample graph images are shown below.  The first was obtained by copying the chart image to the clipboard, the second is a zoomed part of the chart obtained by saving the chart image to a file within the application. Clearly something has gone wrong with the battery voltage scale in the first image.  The emulated battery voltage was around 13.21 V as shown in the second graph, not around 0 V as shown in the first graph.  Regardless, the green trace in both charts is emulated battery Ri as obtained during the discharge profile test.  The orange cursor shows present state of emulated battery charge and the corresponding Ri being emulated. As the test progresses and emulated battery capacity depletes, the orange cursor moves to the right and Ri is dynamically adjusted to match the stored battery profile.  Also note that by the time this test was run I had discovered a Flip check box that allows the Capacity/SOC scale to be, well, flipped.

Figure 27 Capacity and Ri chart during battery depletion emulation

Figure 28 Zoomed portion of Capacity chart

The most noticeable benefit of the BenchVue application in this test was the level of detail it revealed in the inverter current draw.  First, look at the Smart Plug chart showing power consumption of the attached AC load during the test.

Figure 29 Smart Plug power consumption chart

At 1 s sample resolution this chart shows that the charger itself draws about 8.31 W.  When the depleted 10 Ah battery is attached, power draw goes up to 25.0 W, then slowly increases to around 26.8 W.  Now, look at the clamp meter chart.

Figure 30 Clamp meter chart of current draw into 1000 W inverter

This chart was captured during active charging of the 10 Ah battery.  At 1 s sample resolution this chart shows the inverter drawing an average of about 2.82 A from the E36731A acting as a LiFePO₄ battery. This would be about 37 W of input power to the inverter. Now, look at the detail revealed in the BenchVue application.

Figure 31 BenchVue Advanced Battery Application Inverter current draw chart

The sequence of events planned for this emulation test can be clearly seen in this chart.  The emulation was started at 0 s.  At 6.687 s (first current spike) the inverter is powered up. After the inverter stabilizes, the Wi-Fi smart plug is plugged into the inverter at 9.881 s (second current spike). At 25.069 s the battery pack charger is plugged into the smart plug (third current spike).  At 50.582 s the 10 Ah battery is inserted into the pack charger and charging begins. At the time scale of this chart, it is difficult to see what is going on during battery charging other than to note that peak current draw appears to be just under 11.3 A.  Using the zoom and pan tools in the BenchVue application I took a closer look at the charge portion of the chart.  The zoomed in detail is shown next.

Figure 32 Zoomed detail of inverter current draw

It looks like the inverter operates in a pulsed fashion. At 50.582 s the battery is attached and current draw spikes up to 10.5 A, which is the limit set in the Bench Vue battery emulation.  Current then pulses periodically at a rate of about 125 Hz.  Average current draw later in the charge operation was recorded at 2.8 A by the Bench Vue data log, matching the average value recorded by the clamp meter.  I was not aware of any of these operational characteristics before running a battery emulation experiment with the E36731A and Bench Vue software. So, this is a good example to illustrate how using the E36731A can provide actual insights into system behavior.

Evaluation of E36731A programmable slew capability

One feature of the E36731A of particular interest to me relates to a problem with nuisance protection trips in the Load section of the MPPT controller. Steep transient rises in DC load current cause nuisance protection trips.  For example, switching on three parallel connected 10 W LED interior lights trips the MPPT load protection.  Switching them on one at a time does not trip load protection.  Also, connecting the 12 VDC to 5 VDC buck regulator will trip load protection if the input caps on the regulator are not charged.  I am hoping that the ability to adjust current slew in load mode on the E36731A will permit experimental determination of peak slew rate to avoid tripping output protection.  Last year when working without a sophisticated programmable load I used trial and error techniques to estimate the maximum non-tripping current and installed appropriate in-rush current limiting devices on the LED bulbs to prevent nuisance trips. 

What I thought would be an easy experiment to run on the E36731A was actually more complex than expected, and ultimately, it did not provide the answers I was looking for.  Here is what I learned using the E36731A’s variable load slope control.

The specification for variable current slope in load mode provides two maximum slope values.

Figure 33  Maximum Slew rates (p14 of E36731A Data Sheet)

I like to get familiar with a new instrument feature by running exploratory tests.  For the variable current slew feature I set up a range of slew rates and measured them on an oscilloscope.  I set up a bench experiment as shown in the block diagram below.

Figure 34 Bench experiment set up for variable current slew

This arrangement uses the E36731A as a controllable electronic load on the MPPT charge controller.  The charge controller has a 20 A spec limit on its load terminals.  I have two hypotheses about what causes nuisance protection trips on the charge controller load terminals.

1. Excess current flow.

Perhaps inrush currents into low impedance loads briefly exceed the protection threshold, resulting in circuit disconnection.

2. Fast current slew.

Perhaps the protection circuit senses rapid increases in current flow and disconnects the LOAD circuit when a maximum slew rate is exceeded.

With a wide range of current slew rates available on the E36731A and a wide range of settable load currents, both hypotheses can be investigated.  However, the QPX750SP power supply may not behave the same as the 100 Ah LiFePO4 battery.  Using the QPX750SP in place of the battery may introduce new variables into the investigation.  The high sample rate and waveform storage capabilities of the MDO4104-3 oscilloscope provides more insight into system behavior compared to the data logging capability in the E36731A.  The short length of wire from the supply negative terminal to the bare copper connection point provides enough resistance for high transient currents to generate a measurable voltage.

Figure 35 :Three current slew rate settings measured on 'scope

The E36731A was set to draw 20 A from the LOAD terminals for 5 s.  The traces in the image above show the beginning of the current draw at current slew rates of 50 kA/s, 100 kA/s, and 1 MA/s.  None of these tests tripped LOAD protection.  However, they did provide some insight.  It looks to me like I am hitting the current sourcing slope limit of the QPX750SP power supply.  I did two more tests at 10 kA/s and 1 kA/s slew rate (not shown in image above).  With 20 A flowing from the QPX750SP, each waveform reaches a steady state value of about 450 mV drop on the sensing wire.  That means each amp of current produces about 22.5 mV of drop.  Based on this derived value, I calculated the slope of the 10 kA/s and 1 kA/s traces.  Both came out very close to the set values.  However, in the three traces shown in the image above, the measured slopes do not agree with the setting in the E36731A.  The table below shows the setting value vs the measured value on the oscilloscope.

Note also that the 1 MA/s slope is quite distorted.  It begins very steeply, then decreases significantly.  The initial slope on the 1 MA/s trace is about 383 kA/s, but it then drops to around 40 kA/s, like the 100 kA/s trace.  Because slope slew rate settings at 1 and 10 kA/s produced traces that agreed with E36731A settings I believe the E36731A is probably capable of precise current slope control.  The increasing disagreement between setting and measurement as slope was increased to 50 kA/s and beyond may be a limitation within the QPX750SP and not a problem with the E36731A. 

Since I could not generate a protection trip even at 1 MA/s current slew, I tried generating a trip by exceeding the 20 A rating on the LOAD terminals.  Unfortunately, doing so re-exposed a limitation in the E36731A.  With current slew set to 1 MA/s (max setting is 3.7 MA/s), I increased the magnitude of current draw in 1 A increments beyond 20 A. At 22 A both the QPX750SP and the E36731A agreed that 22 A of current was flowing during the 5 s test.  The oscilloscope recorded a voltage drop of about 500 mV, in reasonable agreement with the guideline of 22.5 mV/A. 

When I attempted to go beyond 22 A of current draw the QPX750SP, the Keysight E36731A, and the oscilloscope trace all indicated that no more than 22 A was flowing.  I went up to 26 A on the E36731A, but when the test was executed, all the instruments agreed that only 22 A of current was flowing.  At this point I recalled the 250W power limit in load mode on the E36731A.  It would be helpful if the E36731A gave some indication that the power envelope limited the current, because to the user, it appears the instrument was happy with a current setting that exceeded the envelope.

Since a protection trip was not generated at 22 A and 1 MA/s current slew rate it seems both avenues of exploration have hit instrument spec limits.  There isn’t any point in replacing the QPX750SP with the LiFePO₄ battery because I can’t draw more than 22 A at 13.4 V into the load on the E36731A.  I learned something about the behavior of a couple of bench instruments in this set of experiments, and I learned that whatever causes protection trips on the LOAD terminals of the charge controller is probably current draws greater than 22 A. This outcome represents two partial successes, so I’m happy.

This concludes the evaluation portion of my review.  I found the E36731A to be a very capable electronic load and DC power supply and I found the BenchVue application greatly expanded the capabilities of the instrument for battery profiling and emulation.  In certain tests, I ran into spec limits on the E36731A, but that is fine.  It is up to the user to know the specs of the tools being used and to be aware of how spec limits constrain test capabilities. I would suggest that some sort of warning or indication be presented to the user when instrument settings are likely to exceed or actually do exceed power envelope specs.

Operational and firmware issues discovered during my review.

Firmware issues

During exploration of the E36731A built-in datalogger capabilities I stumbled upon what appeared to be two separate firmware bugs.  When I first stumbled upon these bugs, I made notes about each in my logbook with a plan to later reproduce and fully document each fault.  When I went to reproduce the bugs, I initially was unable to generate the faulty behavior noted in my logbook.  Eventually I discovered that tripping the first bug seems to activate the second bug.  It seems that if the first bug is not tripped, the second bug will not manifest. Most of the details below will only be of interest to Keysight if they try to reproduce the behavior I observed.

1. First bug - Horizontal division setting in data logger mode

I set up a data log with a duration of 30 s and a horizontal time base of 5 s/div.  I was monitoring V1 and I1.  The data log was successfully gathered and saved internally.  I then selected Properties, followed by Navigation to enter a new horizontal time base setting. Using the front panel numeric keypad I entered 10, then pressed Enter.  Screen captures highlighting the area of interest before and after pressing Enter are shown below.

The value in the entry field changed from 10 to 1000.000. Further investigation suggests that any keyed in time base value >9.999 s/div gets mysteriously multiplied by 100. To see if values <9.999 s/div worked correctly, I entered a time base value of 2 s/div and it remained unchanged.  However, once this time per division bug has been activated, the second bug involving marker behavior, becomes active.

1. Second bug - Marker behavior in data logger mode

After returning to the Data Log view, I turned on Markers. Two vertical markers appear, labelled m1 and m2. Marker m1 position is controlled with the Voltage knob and it moves left or right as expected.  Marker m2 position should be controlled by the Current knob. Movement to the right works just fine. Any attempt to move m2 to the left, however, causes marker m2 to jump almost on top of marker m1. Attempting to move m2 to the left of m1 does not work.  If one quickly rotates the Current knob to the left, marker m2 moves a little bit left of m1, sometimes.

Before and after screen captures of m1 and m2 positions are shown below.

The screen capture on the left shows m1 and m2 separated by 2.000 s, with m2 to the right of m1.  I rotated the m2 position control (the Current knob) one click to the left. M2 suddenly jumps to a position nearly on top of m1. The result is shown in the screen capture on the right.  I was unable to find a way to correct this behavior, once activated by tripping bug number 1, other than by executing a power cycle.

Operational issues

1. Inability to locally control output during loss of remote control

I set up a LAN connection between the E36731A and my desktop computer to run a few discharge profiling tests in conjunction with the BenchVue Advanced Battery Test and Emulation application. Twice, immediately after clicking on the Start icon in the BenchVue app, the LAN connection dropped between the PC and the E36731A.  Remote control was therefore disabled.  In both cases an external clamp meter showed 10 A continued to be drawn from the 100 Ah battery. The front panel “On” button was illuminated, but pressing it repeatedly had no effect.  I could not use the front panel “On” button, or the app “stop” button to shut off current draw! The application would not reconnect to the E36731A until I performed a “Refresh Search” operation.  Once connection was reestablished, I was able to use the app “Stop” button to shut down current flow.  The error log showed the following two errors.

Yes, I know the app was not the controller-in-charge, but neither was the instrument front panel the controller-in-charge. My concern here is related to safety.  The buttons that a user would normally activate to shut down a test that has gone sideways did not work.  10 A discharge current is well within operating specs for the battery and the E36731A, so there was no immediate danger, however, the point is, as a user, I temporarily lost operational control over my experiment, and that should not happen.

I don’t know if the LAN issue was OS related within my PC, or something wonky in the BenchVue app, or a firmware issue in the E36731A.  My solution was to switch to a USB connection for all remaining tests.  That solved the lost connection issue for the purposes of this Road Test.

1. Functional improvement recommendations

The Sequencer LIST function in the E36731A can be a useful tool when the user wants to shape the power supply output voltage and current parameters over time, or to do the same with the programmable electronic load part of the instrument.  With 512 steps available in the LIST table detailed waveshapes can be created, as illustrated in the test I performed for this Road Test.  However, supplying a 512-entry table without also providing an efficient user interface to manage the table severely limits the utility of the feature.

My first recommendation is to add an easy-to-use CSV file import and export tool to load and save LIST table data.  512 rows of three element data may not be considered unwieldy by some, but if one must enter the data row-by-row the probability of entry error goes up significantly.  If .CSV store and recall capability is already available, make it easier to find.  Yes, I was able to save data in the LIST table through a save instrument state operation, but that generated a cryptic .sta file that blended a bunch of other instrument state data with the LIST table.

Second, when a LIST is being run, please supply some sort of visual sign on the front panel screen to let the user know which step in the list is being executed.  Ideally, show the step number (step n of x), output voltage, current, and step duration. Could be as simple as highlighting the current row of the LIST table as it is being executed. When I started a LIST run action the LIST table went to gray background and that was it.  There was no sign of what row was being executed, no live display of voltage or current.  I relied on the data log screen and external meters to follow progression through the LIST steps.

Third, to vastly improve editing of the LIST table from the front panel, please add press-and-hold auto repeat for the navigation buttons.   If the user is at the 512^th entry and needs to get to the 1^st entry, for example, the up arrow key needs to be pressed 511 times.

Fourth, and related to the third suggestion, please supply an audible and/or visual indication to the user that the end of the table has been reached.  I often encountered a situation where I needed to change a few values near the bottom of the table and add a few more rows.  The current process requires entry of the value with the numeric keypad followed by a press of the Enter key, then a press of the down arrow cursor key to move to the next row. Even with audible key press beeps activated there is no indication to the user that the end of the table has been reached when the down arrow key is pressed.  Without this alert I found I was overwriting the last row with multiple values when I should have been adding new rows first.  Maybe new rows should auto generate when the user hits the down arrow at the bottom of a LIST table?

Fifth, and this relates to the data logger exported file format, I noticed that rows are sequenced by sample number.  For my application at least, it would be much preferred to sequence rows by elapsed or real time.  The .CSV file holds meta data on sample interval and number of samples, so it should not be difficult to structure the table with elapsed or real time data. I did this once the log file was imported to Excel, but it would be nice if this step could be managed within the instrument.

Thank you Keysight and element14 for providing an E36731A and a BenchVue Advanced Battery Emulation application license to evaluate.  Over the years I have become a fan of Keysight/Agilent/HP instruments.  The E36731A does not disappoint. In my opinion, based on the evaluation conducted for this report, it works, like all other Keysight tools I have used, very well.