Electric Start Investigation


A luxury, once enjoyed, becomes a necessity. - C. Northcote Parkinson

I've gotten spoiled by not having to start my Electric Motion trials bike.  I still love the OSSAs, but I would love them more with an e-start.  Of course, that is not a novel concept.  GasGas and TRS have it, and other manufacturers will eventually too.  

But to me, a conventional e-start system is heavy and complicated for a trials bike.  My recent work with brushless DC motors got me thinking about using the OSSA's 3-phase alternator directly as a starter.  For a primer on BLDC motors, batteries, and controllers see my Electric Motion website: https://www.electricmotiontech.com/home/ev-tech-101 

I can't take credit for the alternator/starter idea, it appears to have been used in the automotive world for quite a while.  Ski-doo is using the concept on their E-TEC direct-injected snowmobiles.

But there are a lot of potential hurdles to overcome.  Whether I end up implementing the idea or not, I'm bound to learn something and will pass the information along here.

Unfortunately, this is not an idea that is generally applicable to other small dirt bikes.  The OSSA's very high-quality overbuilt electrical system designed and manufactured in Japan is the only reason I even considered it a possibility.  

Spoiler alert, I wish I had studied the snowmobile system a bit before commencing these experiments, as I have serious doubts now as to its workability.  Jump to the last section on the Ski-doo SHOT system to see why.

Alternator Suitability

My first task was to identify if the alternator was suitable or not.  From prior work with the OSSA's electrical system, I knew its alternator was a 3-phase machine.  This is the first requirement.  

The photo (at left) below shows the arrangement of the six 3-phase alternator coils (also known as slots), numbered 1 to 6 in the direction of rotation.  The coils are wound with 1 mm diameter wire, which is equivalent to #18 AWG. 

The other (unnumbered) coils provide power directly to the ECU and are not relevant to the experiment or useful as part of a starter motor.  Of course, having the entire stator available for the alternator coils would double the torque.  This may ultimately determine if the project is workable or not.  

The photo (at right) below shows the position of the magnets in the flywheel.  The position of each of the 12 permanent magnets is 30 mechanical degrees apart and marked with an X.  Each adjacent magnet alternates poles like this: 

N-S-N-S-N-S-N-S-N-S-N-S to yield 12 poles (6 pole pairs).

So far, so good.

The 6 stator coils for the 3-phase alternator.

Each magnetic pole marked on the ID with a Sharpie X

Waveform Suitability

My next concern was whether the waveform would be affected by having coils not equally distributed around the stator. 

I spun the engine with a battery-powered drill via the flywheel screw while the spark plug was removed.  I used K-Scan to measure about 300 RPM while spinning.

To observe all three phases simultaneously, I needed to make a 3-phase wye “neutral” connection point from three 1k-ohm, 1/2 W resistors.  This was plugged into the alternator's 3-phase output in place of the stock rectifier/regulator.  The oscilloscope's ground (input common) was connected to the center of the resistor wye.  

The waveforms I recorded are shown below.  Each crude sine wave has a frequency of about 30 Hz (33 ms period) and an amplitude of approximately 4 volts peak-peak when the rotor was spun at approximately 300 RPM.

The equation for motor speed is: RPM = 60 * frequency / magnetic pole-pairs

300 RPM = 60 * 30 Hz / 6 pole-pairs 

Again, this checks out and makes sense.

3-phase alternator output at ~300 RPM.

3-phase alternator output with faster timebase to show detail.

Required Torque

I used my Snap-on dial torque wrench to turn the 250cc motor over by hand via the flywheel retaining screw.  Without the spark plug, it only requires a peak torque of about 20 pound-inch (2.5 Nm).  With the spark plug installed, it was about double that.  I was surprised by these low numbers, although the gearbox was removed during the test so that source of friction was absent.  I also wonder if the slow turning speed may have allowed some air to escape that would otherwise need to have been compressed. 

For reference, the small BLDC motor shown below can produce a peak torque of 250 ounce-inches (1.75 Nm) and a stall torque of 64 ounce-inch (0.45 Nm).

The formula for power using SI units is: Power = Torque * Speed / (30/π)

Let's say it takes 10 Nm and 500 RPM to start the OSSA, that's about 525 watts. 

 Sensorless ESC

I needed a small electronic speed control (ESC) for the next step.  The Chinese hobby electronics marketplace did not disappoint.  All ESCs are designed to drive a BLDC motor from a battery.  My other requirement was that it be capable of sensorless operation.  That is, instead of using Hall sensors or a rotary encoder, the rotor's position is determined via observation of electrical feedback directly from the motor's stator coils.  

I picked an ESC more or less at random via eBay  It was specified to handle 1000 watts with a DC supply of 6 - 24 volts.  The ESC is rated for 50A peak.  So at 12 volts, that's 600 watts of starting power.  

I tested the ESC on a small BLDC motor using a laboratory power supply.  Although this method is generally not a good idea if regeneration is used (most power supplies can't absorb energy as a battery would), I wanted to be able to limit the current for the initial tests.  

The ESC has two controls: a speed (throttle) potentiometer and a Stop-Forward-Reverse switch.  The switch worked well and did not introduce any glitches.  

On my small unloaded test motor at 20 volts, the ESC input current was less than 1.4 amps at max RPM.  Running about as slowly as possible required only 0.1 A.

In searching for technical details, I found a YouTube teardown (embedded below) that turned out to be not at all like what  I found inside mine.  I'll leave it here for the curious because it presents some information that's common to all ESCs.

6-24V, 50A BLDC electronic speed control

Bench testing of ESC with small BLDC motor.  (Motor's Hall sensors not used.)

ESC Circuitry

The next two photos show the ESC with its heat shrink covering and flat aluminum heatsink removed.  I was quite surprised by the high level of integration.  I guess that's the only way something like this can be sold at retail for $14.

The NS51FB9AE is a flash-based microcontroller made by Nuvoton in India.  It is based on the ancient (venerable?) Intel 8051 architecture.  I have a long history with that architecture - it was the first microcontroller I ever programmed in the C language.  

The EG2134 is a 3-phase motor driver chip.  The data sheet is written in Chinese so I'm guessing it must integrate both high-side and low-side MOSFET drivers. 

The other side of the PCB contains a pair of linear voltage regulators, the 78L12 (12-volt) and 78L05 (5-volt).  Since the ESC is specified to operate down to a 6-volt input, I'm a bit uncertain as to why the 78L12 is necessary.  

A large current sensing (shunt) resistor is visible on the incoming negative DC input.  It is marked 0M50” which means 0.50 milliohms.  This is the only provision for measuring current.  Individual phase currents are not measured.

Eighteen HN30N03D  MOSFETs are visible (3 parallel per each of 6 phase switches).  Their logo (CHN) indicates they were manufactured in China.  Rated at 30V and 90A, they have an on-state resistance ranging from 3.1 to 4.5 milliohms depending on the gate drive voltage applied.

Two 470 uF, 35V capacitors are visible hanging off the end of the PCB.  These are known as DC link capacitors.  They provide a low-impedance path for the ripple current created by high-frequency switching of the MOSFETs.

Top side of $14 ESC with heat shrink cover and aluminum heatshink removed. 

Bottom side of $14 ESC

First ESC Test on OSSA

I powered the ESC from an 8-amp bench power supply set to 20VDC.  The rotor (flywheel) oscillated a bit (maybe 15 degrees) but would not rotate.  The current draw was only about 0.5 amps.  Thinking I could perhaps lighten the load, I removed the rotor and key and reinstalled the rotor after lubricating the taper.  The rotor spun easily by hand, but the magnets tend to push it off the taper.  I was afraid it might come loose and reinstalled the M10 Torx screw, but only finger tight.  Unfortunately, the rotation of the rotor tends to tighten the screw and it quickly was tight enough to cause the crankshaft to follow along.  I tried kludging a few different bearings into the mix without success. 

I then thought perhaps the power supply could not supply sufficient surge current.  So I connected an old ~5 Ah lead-acid motorcycle battery with the addition of a Schumacher DSR-108 Ultracapacitor jump starter as well.  But this setup would not cause rotation either.

In retrospect, probably no electronic speed control would work in this application.  The controller probably must run in torque mode.


I did have another electronic speed controller at my disposal.  It is based on the VESC Project.  See: https://vesc-project.com/  Mine is a Chinese implementation made by Flipsky, the FS75100 which is rated for 75V and 100A.

But I was hesitant to even try it as I exploded Flipsy's larger 200A variant when working on a different project.  Because I was still using my 8-amp bench power supply, I figured I would see what I could learn.  The VESC setup software has a semi-automated function to analyze a motor's parameters.  After telling the software that the motor had 12 poles, and I wanted FOC sensorless operation, it returned the following information: 

I assume (the VESC documentation is sparse) these numbers are for an individual coil, even though using an external meter measures a pair of coils in series (wye connection).   My measurements (at the rectifier/regulator connector with the rotor installed) indicated a resistance of 0.78 ohms phase-to-phase (which seems reasonable).  Inductance was 1.1 - 1.9 mH (@ 1 kHz) phase-to-phase, depending on rotor position (again quite reasonable).  Doubling the VESC's 655 uH yields 1.3 mH.  Further, I measured about 800 uH of variation depending on the rotor's position.  This variation is expressed as the Lq-Ld inductance (or inductance in the quadrature axis versus inductance in the direct axis).  Note that with the rotor removed, the stator inductance was about 2.2 mH (@ 1 kHz) phase-to-phase.

During the analysis, the motor rotated back and forth by maybe 30 degrees.  (I have performed that same analysis process on a few different motors and they all made several complete revolutions.)

The good news is that, after the analysis was complete I was able to get the motor to turn approximately 120 degrees (albeit too slowly to ever start the engine).  The bench power supply registered between 5 and 6A during this test.  So let's say 120 watts, but I imagine the peak currents were considerably higher.  And remember, this is with the spark plug removed.  I'm beginning to have concerns the system will be able to produce enough power to start the engine.  The battery will be about 14 volts maximum, and I already needed 20 volts to do anything. 

LiFePO4 Cells

I want to use LiFePO4 cells because they may be charged directly from the OSSA's rectifier/regulator (which is intended for a lead-acid battery charging regimen).

I purchased four 32700-size LiFePO4 cells via eBay.  The total cost for 4 cells was  $28.78.

Each cell is rated 6000 mAh each and weighs 140 grams for a total battery weight of 560 g.   Most importantly, they are rated for a pulse discharge of 48A for 3 seconds.

6000 mAh * 14 volts = 84 Wh energy.

I think it would be possible to use much smaller cells to support the energy requirement if I also use Supercapacitors for the peak current requirement.

Battery Charging

Assuming the starting part even works, I then have to figure out how to recharge the battery.  My first idea is a “snake eating its own tail” concept, where I retain the stock rectifier/regulator for charging.  Ultimately, it would be better to use something programmable like the VESC and perform a regen of sorts.  That would be way more work, though.  One hurdle at a time.

Supercapacitors / Ultracapacitors

Supercapacitor and Ultracapacitor are just two different tradenames for the same technology.   

I ordered a Supercapacitor bank for experimentation via AliExpress for $16.37, delivered.  The bank is rated at 16V and 16.6 farads!  It comprises six 100F, 2.7V capacitors connected in series and can support a 50 - 60A discharge current.

The bank also contains some active circuitry to equalize the charge among the six individual capacitors.

gdchpcap.com capacitor bank rated at 16 volts and 16.6 farads

Supercapacitor Bank Energy Storage

The total energy available is minuscule compared to a battery of similar volume.  The formula for energy stored in a capacitor is: U = 0.5 *  C  * V² 

U = 0.5 * 16.6 * 14.6² = 1769 joules

There are 3600 joules to the Wh, so only 0.49 Wh of energy is stored.

Below is a discharge graph I produced for the unit purchased.  It was charged to 14.6V (same as a 4-cell LiFePO4 battery) and discharged at a constant 10-amp rate (max I could achieve with the test instrument) down to 0.5 volts.  This took about 30 seconds.

The bank released 0.61 Wh (more than calculated, so the capacitance must be greater than the rated 16.6F).

Supercapacitor Balancing Circuit

Capacitors connected in series do not necessarily distribute voltage equally among themselves, so a balancing circuit is required for safety and to circumvent accelerated aging.  The adjacent photo shows the components required for each capacitor.

U1 is marked “CNVL” which indicates the XC61CN2702MR, a 2.7-volt precision voltage detector in a SOT-23 package.

Q1 is unmarked.  Other balancing circuits use a silicon NPN transistor.

R3 (marked 3002) is 30k ohms, 1%

R1 and R2 (marked 4R7) are 4.7 ohms.  They are connected in parallel.

I did not bother to trace the circuit (there's a conformal coating on the PCB that insulates the components) but think what I've drawn below is likely correct.  The transistor does not turn on until there is 2.7 volts across the Supercapacitor.

Likely balancing circuit

Balancing components required for each Supercapacitor


Back EMF at high engine speeds.  The ESC's power MOSFETs must be capable of withstanding the voltage being produced even in their off state.

Stator heat dissipation from a high-current start.  I'm guessing the OSSA alternator can deliver roughly 10A continuously.  So a 50A pulse for a few seconds should not pose a problem.

But I don't want to burn up an essentially unobtainable stator.  Although I could possibly have it rewound or rewind it myself, that would be a big PITA.

For testing, I can decrease the engine's pumping losses with the throttle wide open.  

The extra inertia of my added flywheel weight probably makes the problem more difficult. 

Ideally, the wires from the controller to the alternator would be as short and thick as possible.  This means butchering the stock wiring harness.  I don't mind making reversible changes to the bike, but building a completely new wiring harness is a lot of work.

Ski-doo SHOT System

The SHOT system is a Ski-doo (Rotax) design to start a snowmobile using the alternator as a starter motor.  Its design highlights several things I discovered to be impediments to doing something similar with the OSSA. 


Credit: Ski-Doo.  SHOT System overview page from the Ski-Doo electrical service manual.

A Modicum of Success

Eventually, I was able to get the rotor to turn slowly (several seconds per revolution) making complete revolutions jerkily.  But I needed ten Li-ion cells in series (40 volts).  The cells were rated for a peak discharge of 20A.  The VESC was configured for FOC (field-oriented control) mode.  After a few revolutions, I detected a smell that was probably insulation getting hot.  I removed the rotor to feel the coils - they were warm but not uncomfortable to touch.