Most signal conditioners don't have particularly discriminating filters. They are designed to take out PWM artifacts, which are in the kiloHertz range and above. You are going to have a huge noise contribution due to the feeder.
One of the issues you most likely have with the filter in the plc program is update rate. To effectively filter a signal you need to be sampling faster than the signal. The plc scan and analog update time may be getting in the way of this. You might also want to try using the filters on the input module itself. These run much faster and can do a much better job of attenuating noise.
If that doesn't work take a look
here. Go under the product name index and look at the GSAF-10. I've used these in tension control applications in the past to get rid of the vibration of a load cell roll. It is a 9-pole filter and you can specify the corner frequency. It produces some phase delay but for your application that shouldn't matter much. I ordered a 15 Hz corner frequency and could barely see a 60 Hz full voltage input on the output side. I recommend the Bessel filter as it won't ring given a step input.
If you really want the Cadillac solution look at
this. Take the time to view the videos; they are rather impressive and very applicable to what you are doing. Specifically look
here. there is also a version for the 1769 I/O format. It will perform the rate of change calculation for you. They are a bit pricey but so is your time.
Are you determining feed rate but looking at the change in weight?
I also still recommend you implement a feed forward term if you haven't already. This may end up being somewhat non-linear but anything you can add to the output that the PID doesn't need to generate will help, especially when you change feedrate and start up.
Keith