Low-pass Filter

I am currently working on a slitter/rewinder for plastics. The rewind tension control is loadcell based. We are trying to run about 2lb of tension out to 2000 FPM. My tension signal is getting swamped by mechanical noise. I don't have a full frequency breakdown of what I am seeing but I know for sure I have a VERY strong component at about 60- 70Hz as well as a weaker component at about 110 Hz. I'm sure I am also seeing some sideframe vibrations at higher frequencies that just show up as a block on the recorder I am using.

I am looking for a low-pass filter to get rid of the mechanical noise. I was thinking something with a 5 - 10Hz cutoff. I have three tension channels so a multi-channel unit would be OK, although multiple single-channel units are fine, too. I would like something free-standing (DIN mount, stand-alone board, etc) but I would consider rack-mount solutions.

Does anyone have a supplier for something like this?

Thanks,
Keith

In my old age, I have trouble remembering my name (which is Keith), so it may be a while trying to come up with a filter manufacturer name here.

I ran into something that may be similar a long time ago on some portable equipment. We had strong interference around 67, mild at 110, 118, 123, and 127, and strong at 156, 167, and 192, in memory serves me here.

What type of wire are you using?

We switched to high quality shielded high-twist multiple pairs in some areas, and spent a lot of tyme with some type of reflectometer looking for the interference origin.

We used a Bird Model 43 wattmeter with a special slug that had a 6" flex antenna. Don't remember it's name or number.

Biggest problem was bad electrical connections in plant, poor grounds, and some really strange ground loops. Including one or two shielded cables grounded at both ends.

regards.....casey (KC-Keith Curtis)

I've implemented load cell based tension control on a couple of tissue rewinders (up to 3000 fpm). Frame stiffening and roller balancing play a big part in reducing "noise". In my case, perforating the tissue sends shock waves through the web, adding to the "noise". I'm using ControlLogix and find that the LPF function block does the job well.

I know this has been posted before

For a simple low pass filter

y=A*y(n-1)+(1-A)*x
A=exp(-2*PI*Hz*T)
Hz is the corner frequency.
T is the update time.

A two pole Butterworth filter is better.
ftp://ftp.deltacompsys.com/public/PDF/Mathcad - Butterworth.pdf

The Butterworth filter as a much steeper cutoff at the desired frequency. The downside is that is requires two extra mulitplies and adds. I would use the Butterworth filter.

A low pass filter isnt hard to make

Thanks for the info, guys.

The rolls on the machine are balanced to < 0.05 oz-in, which I believe works out to an ISO balance grade of between G1 and G2 for my roll weight and speed. How tight should I go with balance? I would have thought that level of balance would be pretty good.
We've already stiffened the loadcell shaft. We are considering trying to isolate the roll shaft from the loadcells with rubber isolators. But I don't think the frequency is high enough for that to help.
We are tied into using the Eurotherm LINK drive system on this job. So I won't be able to use the low pass filter in a CLX processor.
Has anyone tried carbon fiber roll shafts in cases like this?

Thanks for the thoughts.

Keith

Thanks for the reply, Peter.

The drive system I am using has a single pole low pass software filter in it. I tried to use it but it didn't help much. I think the problem is the period of the major noise component (about 15 msec) matches up with the analog sample rate of the drive (15 msec). So I'm probably getting some pretty serious aliasing on the input to the filter. I have used the filter equation you posted quite frequently. It usually works out pretty well for me.

Thanks for posting, Ron. I tried the resistor/cap approach and I may return to that yet. I think I messed myself up with the resistance values I used. I couldn't get the voltage after the filter high enough. I think my filter resistor value was too high and i made a voltage divider with the drive input resistance. I will get my hands on some additional resistor values and try some other combinations.

One of the reasons I was looking for a stand-alone filter is that it would be a cleaner install that way. I can probably get away with hanging a resistor/cap combo somewhere. But my customer will nmot look kindly on something I toss together on perf board and wire wrap.

Thanks,
Keith

I wonder if there is a market for this

Make a small analog filter board using a chip. Companies like maximum have chips that implement a 4 or 5 pole filter with a very steep cutoff. There are many on-line PCB layout websites that can send you a small board in a couple days.

I am surprised one cannot buy an off the shelf filter. I did a quick search but all I found are sites telling one how to make the Butterworth or Elliptical filters.

See if they have something you may use:
http://www.alligatortech.com/default_low_pass_filter.htm

I have more but have to look them up.

Thanks for the link, Ron. I found Alligator last night. But they were about the only ones I found. There were a few others I came across but all the others made filters that looked like lab test equipment not industrial stuff. Alligator has several premade designs with cut-offs at 1kHz and above. They will make you a custom cut-off at anything down to 2Hz. I will probably check into them to get some idea of availability.
Like Peter, I found several IC manufacturers that make filter chips but very few pre-made filter boards.

Thanks again,

Keith

I ordered three filters from Alligator today (GSAF-10/L-15). They are 9-pole Bessel filters with a corner frequency of 15 Hz. I'm a little worried about the linear phase delay causing me some trouble at lower frequencies but we'll see.

I'll let everyone know how this turns out.

Keith

So how much did they cost?
Why 9 poles? That is a lot of poles?
Why Bessel?

The phase delay is a killer. We have 4 pole Butterworth filter built into our newer products, but I don't like to use them for feedback because of the phase delay. The filters are mainly used for HMI valves and for filtering analog position or velocity references.

I would be embarrassed to have a 15msec sample rate on my product. That isn't fast enough. We over sample every 122 microseconds or 8192 a second on each analog channel. That provides enough data where the software filters can di their job. We use the filters on a chip for anti-aliasing filters. These anti-aliasing filters are obviously up in the 4 khz range.

The filters are $445 a piece. They are really marketed as stand-alone anti-aliasing filters. Their 'standard' offerings are designed around several center frequencies starting at 1kHz and going up to 50kHz. But they will design around any center frequency from 1Hz up to 50 kHz.

I was talked into the Bessel filter by the tech sales guy at Alligator. I told him I was using this in a control application and he said the Bessel is a better bet because it has a uniform group delay and won't overshoot on a step input. I don't know enough about filtering to argue the point and I was under the gun to get something ordered so we get the filters here. But quick ordering won't do me much good if what I get doesn't work for me.

I have to think the high pole count is just to get a steeper roll-off. The specs say the filter will attenuate at -27dB/octave. That will give me -54dB attenuation at the frequency I am having trouble with, which is more than I need.

Yes, a 15msec sample rate isn't real hot. But it is just a general purpose analog input into an AC FOC drive. It wasn't specifically designed to handle high frequency signals. In reality, my tension signal shouldn't contain the frequency components that it does. The web can't really change tension that fast and the drive drive can't respond to frequncies that high due to the torque to inertia ratio. But the low sample rate does preclude the use of effective software filtering, as you said.

So, if you don't like using filtering in the feedback path due to phase delay, how do you deal with objectionable mechanical frequency components that might be in the feedback signal?

Keith

kamenges said:

The filters are $445 a piece.

Click to expand...

Wow, pricey. I will pass that on to our analog designers.

kamenges said:

I was talked into the Bessel filter by the tech sales guy at Alligator. I told him I was using this in a control application and he said the Bessel is a better bet because it has a uniform group delay and won't overshoot on a step input. I don't know enough about filtering to argue the point and I was under the gun to get something ordered so we get the filters here. But quick ordering won't do me much good if what I get doesn't work for me.

Click to expand...

The tech guy is right. I think you made a good choice. I was just curious because this is way past the scope of this forum yet it is good that everyone knows these devices are available.

kamenges said:

I have to think the high pole count is just to get a steeper roll-off. The specs say the filter will attenuate at -27dB/octave. That will give me -54dB attenuation at the frequency I am having trouble with, which is more than I need.

Click to expand...

Yes, filters are a trade off. If you get too much filtering you may also get too much phase delay.

kamenges said:

So, if you don't like using filtering in the feedback path due to phase delay, how do you deal with objectionable mechanical frequency components that might be in the feedback signal?

Keith

Click to expand...

I tell the mechanical engineer to fix the mechanics so the natural frequency or resonances are higher. You can't fix everything in software or with filters.

Originally posted by Peter Nachtway:
Wow, pricey. I will pass that on to our analog designers.

Click to expand...

I thought that was a little steep, too (pardon the filter pun). But they are one of the very few players in this type of package. And, ultimately, if it works, it's worth the price. But I would say there is some serious room for competition in this market.

Originally posted by Peter Nachtway:
I tell the mechanical engineer to fix the mechanics so the natural frequency or resonances are higher. You can't fix everything in software or with filters.

Click to expand...

I may have to print that one out and post it on my wall. The mechanical guys around here don't seem to think that way. Maybe a 3ft X 5ft message board with that quote might get the idea across.

Thanks for the help.
Keith