[Stiive]

Hi,
I'm about to implement a 3-phase IGBT based sensorless DTC-SVM induction motor drive and was wondering if someone here (or know of a project somewhere) has already done similar. Would be great to compare schematics and/or code with someone who's already got a working system.

I have got my C code running my induction motor flawlessly using the Induction motor on SimPowerSystems. To reduce costs, my plan is to use a couple of fast micro controllers to run the code (one for d-q transforms and one for lookup) as i have eliminated all FFT and therefore hopefully will be fine without a DSP.

I have some 600V 600A IGBTs that i'll probably end up using, depending on the final design.

I am aware of the Tumaniko KiwiAC project, but that uses FOC which obviously has different requirements in hardware/software. However, FOC would still be useful to compare notes on areas such as; size of capacitor bank, snubbers, water cooled heatsink CFD designs, current/voltage sensor, procurement etc

Cheers

[bjfreeman]

long thread few years
Converting 2001 Passat

[Stiive]

Thanks for the link.
I remember reading about this project a few years back. I will give him a PM.

Anyone else?

[JRoque]

Hi. Forum member jhuebner is also working on the Tumanako project and might be able to help cross-checking your work.

How are you passing data between the two micros? Is that link going to be quick enough? I've also wondered if one micro with enough local memory can do both main code and lookup tables. They make them now with up to 256Kb and more for higher bit counts.

Have you looked into Microchip's DSPic chips? Those are not that expensive and combine both needs into one die.

Last but not least, are you planing on contributing your design to forum members or commercialize it?

JR

[Stiive]

Originally Posted by JRoque

Hi. Forum member jhuebner is also working on the Tumanako project and might be able to help cross-checking your work.

How are you passing data between the two micros? Is that link going to be quick enough? I've also wondered if one micro with enough local memory can do both main code and lookup tables. They make them now with up to 256Kb and more for higher bit counts.

Have you looked into Microchip's DSPic chips? Those are not that expensive and combine both needs into one die.

Last but not least, are you planing on contributing your design to forum members or commercialize it?

JR

JR,
Thanks for your response.
I was looking at using 2 80mhz MCU demo boards. Alot of the data can be sent in parallel using a few ports each and would just be the output of the hysteresis controllers (0-4), flux sector (0-5) and angle (if you are familiar with DTC operation). Other information for debugging or advanced features can be sent in serial.
Because the control is SVM, this is perfect for two processors (or dual core), with the first CPU computing the preview vector, and the second implementing the smaller vectors. Infact i originally started design using the 8core propeller MCU, but hated the language so scrapped that.

The DSPic is definitely the path I will go down once I have proven the code, and finalised PCB design. If i could find a cheap DSPic 70MIPS demo board then maybe i would use that straight up.

Not gonna be selling these, don't own the right to DTC patents. Maybe I would give the design upon completion of a electrical theory and safety test lol. Don't want anyone killing themselves over a schematic i've given them.

If there's interest, might create a youtube series on the build. Have already done the model in simulink (started with blocks then replaced with code), Solidworks design of the finished product, CFD on the water-cooled heatsink, and eagle PCB surface mount design of the finished board if it all works as expected.

[bjfreeman]

might look at the STM3F4 series.

[bjfreeman]

Originally Posted by Stiive

JR,
Thanks for your response.
I was looking at using 2 80mhz MCU demo boards. Alot of the data can be sent in parallel using a few ports each and would just be the output of the hysteresis controllers (0-4), flux sector (0-5) and angle (if you are familiar with DTC operation). Other information for debugging or advanced features can be sent in serial.

your already started with your design so this is more of how you arrived at you design.
You stated you did a processor without DSP, is that the reason you need 2 mcu.
I have been using a 80 Mhz DSP 56F803 for years and it handles the job of drive management.

Because the control is SVM, this is perfect for two processors (or dual core), with the first CPU computing the preview vector, and the second implementing the smaller vectors. Infact i originally started design using the 8core propeller MCU, but hated the language so scrapped that.

The DSPic is definitely the path I will go down once I have proven the code, and finalised PCB design. If i could find a cheap DSPic 70MIPS demo board then maybe i would use that straight up.

The name infers a DSP, are you deciding a DSP is necessary?
I have found that I needed to optimize the machine code macros created by the Cross compiler, so my decision was based on clock cycles for instructions.
Most chips range from $8-$20 in single item orders.
Most dev boards Range from $10-$40, manufacture proto boards being the most inexpensive and way below the Cost of production.

Not gonna be selling these, don't own the right to DTC patents. Maybe I would give the design upon completion of a electrical theory and safety test lol. Don't want anyone killing themselves over a schematic i've given them.

If there's interest, might create a youtube series on the build. Have already done the model in simulink (started with blocks then replaced with code), Solidworks design of the finished product, CFD on the water-cooled heatsink, and eagle PCB surface mount design of the finished board if it all works as expected.

have you thought about adding the modeling into the actual MCU as a module.
there are Graphic LCD(3.5 inch) that are inexpensive ($30) that can be driven from a MCU that has enough ports or using Ic2. or you can use a graphic library to generate the RGB an sync to output to a computer monitor.
if you find that module slows up the drive management you can off load it to a second mcu.
The stm32f4 discovery board is about $20.

[Tesseract]

Originally Posted by Stiive

Hi,
I'm about to implement a 3-phase IGBT based sensorless DTC-SVM induction motor drive and was wondering if someone here (or know of a project somewhere) has already done similar. Would be great to compare schematics and/or code with someone who's already got a working system.

I have got my C code running my induction motor flawlessly using the Induction motor on SimPowerSystems. ...

Ambitious project with some heavy duty simulation and code generation firepower behind it. Unfortunately, no one else in the DIY scene that I am aware of has the $25k worth of software you seem to have at your disposal to tinker around with this stuff. I certainly don't...

That said, I did a lot of research on DTC vs. FOC about a year and concluded that DTC offers no compelling advantage over FOC. Both schemes are kind of kludgy in my opinion (way too much reliance on "estimation" and "modeling" of critical parameters) but DTC is a variable switching frequency scheme by nature and that presents some monumental obstacles from a hardware design standpoint. For example, the amount of DC link capacitance is inversely proportional to switching frequency.

Also, the flux and torque estimator for DTC needs to know the stator voltages and currents. Obtaining the current is relatively straightforward, but measuring the voltage on each phase without distorting it from filtering while also needing to heavily filter out the inevitable noise and ringing from the waveform is exceptionally difficult.

I thought DTC was the greatest thing since sliced bread at one time... not so much anymore.

[Stiive]

Thanks for your post

Originally Posted by Tesseract

That said, I did a lot of research on DTC vs. FOC about a year and concluded that DTC offers no compelling advantage over FOC. Both schemes are kind of kludgy in my opinion (way too much reliance on "estimation" and "modeling" of critical parameters)

DTC requires only the knowledge of stator resistance - unless your doing some advanced sensorless speed estimation

Originally Posted by Tesseract

DTC is a variable switching frequency scheme by nature and that presents some monumental obstacles from a hardware design standpoint. For example, the amount of DC link capacitance is inversely proportional to switching frequency.

Excactly why i was hoping someone else had done this before :P My plan is to just over compensate. With my model I can find the usual switching frequency bandwidth for the motor, and work from there.

Originally Posted by Tesseract

Also, the flux and torque estimator for DTC needs to know the stator voltages and currents. Obtaining the current is relatively straightforward, but measuring the voltage on each phase without distorting it from filtering while also needing to heavily filter out the inevitable noise and ringing from the waveform is exceptionally difficult.

Yup, spot on. The voltage sensors have been causing me headaches... Might just keep it simple in the end with an opamp circuit and see how it goes. I am also working on some code to hopefully eliminate this voltage requirement completely (or atleast reduce its sensitivity).

Originally Posted by Tesseract

I thought DTC was the greatest thing since sliced bread at one time... not so much anymore.

Unfortunately I need DTC for my project otherwise I prob woulda just modified an industrial VFD, that seems the easiest. I have a Rockwell PowerFlex 7.5kW drive that would have been great to hack.

That being said, I still think DTC is easier than FOC for the DIYer - but time will tell!

[Tesseract]

Originally Posted by Stiive

DTC requires only the knowledge of stator resistance - unless your doing some advanced sensorless speed estimation

I don't know about DTC only needing to know the stator resistance - there are many variations on the original scheme, you know, including marrying it with SVM - but either way the stator resistance changes with temperature and that was what I was getting at when I wrote that it's performance depended heavily on modeling/estimation.

Parameter shift causes all sorts of headaches for both DTC and FOC schemes, so I'm not singling out DTC here.