Commanded Spindle Speed Is Incorrect

I have just wired up a HuanYang VFD to a 2.2kW spindle and have LinuxCNC talking to the VFD via its analog input. I have PDM output on pin 14 of my parallel port. This is wired to a Gecko G540 which filters the PDM to create an analog voltage. The voltage is then fed to the VFD. I have setup my hal file as below:

This all works to the extent that I can get the spindle running at speeds varying between 0 and 24,000 RPM. However, the actual speed does not match the speed I command. It is off significantly in some cases.

A few notes:

• The VFD is correctly scaling speed according to the input voltage at the terminals. The error is in the voltage ouput from LinuxCNC/Gecko. For example, I command S12000 which should be half speed according to my scale settings. That should present 5V (0-10V), yet I see 7.06V.
• It's not a problem of resolution as I am able to step the spindle at ~ 100 RPM increments under manual control.
• Looking at the output on HAL Oscilloscope things seem ok.

Any thoughts?

I would try PWM instead of PDM
and see if it improves the linearity
(at the expense of more ripple)

I'm guessing that the G540 uses a slow OPTO coupler
which will have very unsymmetrical on and off times
causing a large amount of non-linearity.

Thanks for the response! I actually started off with PWM and changed to PDM in hopes of it solving some problems. However, I find they operate identically (with respect to the problem I am having). The error seems to be right about 140% of my set point in a series of tests I performed. Is there a way to view the commanded value in HAL meter as a float? Meaning, I could verify that S12000 is indeed 0.5 and S18000 is indeed 0.75?

What I would do is this:
Set the PWMGEN to PWM mode, set the spindle speed to 12000 and then look at pin 14 with HALScope. If the PWM waveform has 50% duty cycle the trouble is elsewhere.

I'll try that again, but I believe I verified it looked correct at 50%. As I recall it was hard to be certain since the zoom feature would not allow me to zoom in on a single period. However, for the amount of error I am getting it should be quite obvious from the waveform.

I tend to agree with your initial thoughts regarding the optocoupler. I was thinking I'd try to lower the PWM frequency so that this effect was less evident. I was going to blindly adjust this but I'd rather have some understanding of how this relates to the base frequency so that I may pick a logical value for the frequency.

Also, if not specified, what is the default frequency? I thought this might be stated in man pwmgen but I was mistaken. Most of the spindle speed examples don't actually specify frequency.

man pwmgen (in a terminal window) should help with most of these questions...

You seem to be correct in your initial assumption. If I reduce the PWM frequency from 10k to 1k I get a much better result. I am going to leave it there. I lose some resolution but it's perfectly fine for what I need. In addition, I also enabled dithering which seems to help gain back some of that lost resolution.

In the quest for perfection I took the small DC offset out of the picture by adjusting the offset parameter for that PWM generator. The VFD reads EXACTLY what I command now.