Marlin 2.1.1 Model Predictive Control has deviation of -3 to +1 degrees

Hi,

I tried the Marlin 2.1.1 version with MPC for the extruders.
The tuning was successful, but there is a constant error for all set points, which is also periodic. The waveform is attached.

I have tried tuning multiple times with the same results.
I am using PT100 with 1k resistor as potential divider and a 60W cartridge heater.
Kindly suggest any possible solutions.
Thanks in advance!

My MPC deviation was about 1/4 of what you're seeing. About the same a PID.

Yes, I expected it to be better, but it's slightly worse than PID. However, one thing I notice is, in PID, the errors are random. In MPC, the error is periodic. I have a second extruder which behaves exactly the same.

Edited 1 time(s). Last edit at 12/03/2022 03:53AM by srinidhi.

I suspect your 60W heater is not 60W

Unplug the heater, measure its resistance

then do some simple math



to calculate the real wattage (P)

Edited 3 time(s). Last edit at 12/03/2022 01:20AM by Dust.

Will try this and update, thanks

make sure to touch your probes together to measure the error Resistance of the meter and probe wires then subtract that value from the heater resistance measured

Edited 1 time(s). Last edit at 12/03/2022 01:29AM by Dust.

Hi, I found the resistance of the heater to be around 11 ohms.
Power = 24*24/11 = 53 Watts
This value has slightly improved the performance, the errors are not periodic, and have a deviation of around -2 to up to +2 in some cases. Is it lesser than this usually?
I am thinking of iterating with different values of power, and finalize the one which gives better performance.

The MPC M306 T routine places the nozzle at 1 mm from the bed by default, and then engages the part cooling fan to calculate the heat loss during actual printing. I found that 1 mm was NOT close enough, and set the distance to 0.1mm in configuration.h (my typical first layer height).

It made a huge difference, at least for my set-up.

#define MPC_TUNING_POS { X_CENTER, Y_CENTER, 0.1f } // (mm) M306 Autotuning position, ideally bed center at first layer height.

Instead of:
#define MPC_TUNING_POS { X_CENTER, Y_CENTER, 1.0f } // (mm) M306 Autotuning position, ideally bed center at first layer height.

Also be aware that many control boards have very bad ADC (analog to digital converter) designs, and they lie like a dog on a sun warmed rug. It is very unlikely for a heat block to swing many degrees over a few seconds.

In addition, a PT100 RTD is possibly the worst sensor for this application. A 12 bit ADC can resolve, at best, 4096 steps, or 0.122C over a 0C to 500C range (about 0.8mV with a 3.3 Vref), while the voltage from a PT100 with a 1K exciter is only 0.000669 volts PER DEGREE. The temperature readings will appear to bounce wildly, because the ADC hasn't the resolution to do anything else.

I use a PT1000 RTD (10 times more sensitive), with an amplifier to linearize and expand the RTD signal to cover the entire ADC input, and I do run exclusively with MPC. But you may just be expecting resolution and stability that your set-up just isn't capable of.

Edited 2 time(s). Last edit at 12/05/2022 12:38PM by rq3.

Hi, thanks for the inputs!
I'll try out changing the centre position. You may be right regarding the resolution, but will give it a shot and update.