Yet another Lebowski + Honda IMA inverter & step-by-step

[kiwifiat]

@kiwifiat Sorry if this a question with an obvious answer.
In this thread I see a newer version of the Lebowski board you designed that looks to be designed to drop directly onto the standoffs of the IMA inverter.
looking through other threads the best I have been able to find is the small Bobc board style. Not the one like is used in this thread.

The board has support for the resolver fitted to the HSG, just connect and tune. The board gerbers haven't been released yet as there is a revision in the works that makes connecting to the outside world less of a hassle. The current design is meant to connect to the existing oem 20 pin socket, seemed like a good idea at the time, but it is ridiculously time consuming.

 

[kiwifiat]

Makes me think it's some fundamental property of motors rather than an actual controller issue and that the problem is just masked rather than fixed by "tuning". Possibly at the expense of performance in each region.

I wonder if @Lebowski has any insight into this.

IMO the issue is that HALL sensors are crap, especially when they are reliant on magnetic fields and they are placed in a lightning storm of magnetic fields that vary with phase current. The best solution was that proposed and solved by ES member Bertie using optical sensors that are impervious to changes in magnetic fields. That said I have had zero issues with the Lebowski controller with transition from sensored to sensorless operation on my old scooter that was fitted with a gen1 QS273 motor and HALL sensors. Those that I am aware of who were having transition problems traced the issues to sub-optimal alignment of the magnet on their RLS encoder fitted motors.

 

[Lebowski]

Hall sensors are indeed finecky… the best is when they are external to the motor and do NOT see any magnetic field from the coils. This is the case I think I used for my controller. The case where the sensors see the field of the coils is fundamentally different…

It would cool to take the time to make a useable-for-everyone version of the ‘sensorless from standstill’ algorithm as this is far superior. Who knows, hopefully I have time to get back into this in the near future.

 

[mxlemming]

Hall sensors are indeed finecky… the best is when they are external to the motor and do NOT see any magnetic field from the coils. This is the case I think I used for my controller. The case where the sensors see the field of the coils is fundamentally different…

It would cool to take the time to make a useable-for-everyone version of the ‘sensorless from standstill’ algorithm as this is far superior. Who knows, hopefully I have time to get back into this in the near future.

Does your sensorless from standstill work for all motors? Or does it require a degree of salience?

I read and understood the paper (I think), but noted that you have orders of magnitude more sensitivity in your hardware than normal FOC controllers due to the high pas filtering opamps. In my world, I created SFS that works with many motors but relies on running an angle error correction based on the Id Iq currents induced by VdVq injection. It is very very good for some motors and totally hopeless with others.

 

[Lebowski]

It requires some small degree of saliency cause thats what the principle is based on. I added the opamps to increase sensitivity (amplification) but also the be able to detect saliency while applying large torque producing phase currents (filtering).

You can do without the opamps when you have audio quality AD converters, but I want to use a microcontroller with the ADC’s inside. Also then you need lots more processing power to do all the filtering in the digital domain… and this is all assuming there’s no harmful aliasing.

 

[beggernator]

The board has support for the resolver fitted to the HSG, just connect and tune. The board gerbers haven't been released yet as there is a revision in the works that makes connecting to the outside world less of a hassle. The current design is meant to connect to the existing oem 20 pin socket, seemed like a good idea at the time, but it is ridiculously time consuming.

Awsome news.

Is there any sort of expected ETA on the revision?
I understand it's a hobby project you probably work on as and when in free time so no rush. Just would be nice to know if I'm shelving it temporarily for a few months or going full cold storage for year +.

As a thought in the connector location, you could place a circular mill spec style connector on ether the back or the side of the case. Some difficulty with removing the lid I suppose. Other option is just cutting the OEM connector off the board and mounting a new connector in its place. Side mount bolts through the PCB provided the traces under it are non essential. Still a lot of messing around but less then re useing the OEM one.

I spose if the new board revision has THT wire breakout points and a provision for an on board connector that gives the most options for least effort.

Thanks again to everyone's combined effort here. It's great to see what a community can produce when It works together.

 

[kiwifiat]

I will be using a 23 pin Ampseal header. Early April as a wild guess for an ETA. I still have one populated V1 board available from the initial run of ten, pm if you are interested.

Cutting the oem socket out is painful as it is screwed in from below the PCB with no access to the screw heads. I cut one out with a dremel and it created a hell of a mess.

 

[mxlemming]

It requires some small degree of saliency cause thats what the principle is based on. I added the opamps to increase sensitivity (amplification) but also the be able to detect saliency while applying large torque producing phase currents (filtering).

You can do without the opamps when you have audio quality AD converters, but I want to use a microcontroller with the ADC’s inside. Also then you need lots more processing power to do all the filtering in the digital domain… and this is all assuming there’s no harmful aliasing.

I got it working with pretty big phase current (160A). Had about 2/3 my weight on the bike saddle. I'm 90kg. Unfortunately it's only silent because I'm running 35khz pwm so half that from positive and negative injection is 17.5kHz

Something worth noting is that if you construct your injection and reading right you find that higher current stabilises the lock in, whereas other combinations are unstable with higher current.

Injecting equally in d and q axis and choosing the polarity of the q to be counter to the current is stable. If q injection is towards the current, then higher q current causes q saturation which ultimately removes the apparent saliency.

Iirc from your paper your reconstructing an entire circle from the injection, and you can see the saliency from the ovalness (very crudely transcribed). As you apply more and more torque, the circle gets less pointy on the high current q axis side. The saturation shift with current enables the 180 degree disambiguation... Effectively I just chose a single point on that circle that seems relatively unaffected by current and track that. Maybe that works well on your opamp free hardware?

 

[Lebowski]

For my testing I used a drill

The idea was to be able to run the big screw fully into and out of the block of wood, sensorless and from standstill. So even stopping for the last cm, and then re-starting from sensorless. The motor was pretty small

I ran it at 50A phase current. It worked flawlessly. Never tried it on a bike as here the law is too strict now for home brew.

From your short description I don’t get what you’re trying to do…