Group Actions
Engagement
  • Author Author: Jan Cumps
  • Date Created: Date Created
  • Views 424 views
  • Likes 5 likes
  • Comments 25 comments
Related
Recommended

PID temperature controller for the EasyL1105 MSPM0 board - Pt. 3: PWM

Jan Cumps
PID temperature controller for the EasyL1105 MSPM0 board - Pt. 3: PWM

  designed a development kit for the recent Texas Instruments MSPM0 microcontroller series. 
This 4 part blog series documents the steps to design a PID temperature controller. Part 3: add PWM to generate a PID controlled output.

image
(post that introduces the kit)

Goal of this 3nd post

not a goal of this post: have the PID regulation working 100%.

Set up PWM SysConfig

The code uses timer TIMG1, channel 0, to drive PA26. There's no interrupt involved. The duty cycle gets adjusted in the regulation loop.

At this point, I have set the PWM period count to 65535, in an attempt to have the output range identical (but 32768 offset) to the input. 

image

image

Code

note: this design doesn't regulate perfectly yet. At this point in the blog series, all modules are in place and they are tied to the PID. But finetuning PID parameters, input, output, ADC and PWM settings is for post 4.

void perform_pwm() {
    DL_TimerG_setCaptureCompareValue(PWM_0_INST, i32_Output_PID + 32768,
        DL_TIMER_CC_0_INDEX); // update ccr0 value  
}


int main(void) {
    SYSCFG_DL_init();

    // /* timer 5 interrupt ticks per second */ 
    // /* Enable Timer0 NVIC */
    NVIC_EnableIRQ(TIMER_0_INST_INT_IRQN);

    NVIC_EnableIRQ(ADC12_0_INST_INT_IRQN);
    gCheckADC = false;

	/* Initialize the parameters of PID */
	Initialize_PID_Parameter();

    /* Start PWM */
    DL_TimerG_startCounter(PWM_0_INST);

    /* Start Timer counting */
    DL_TimerG_startCounter(TIMER_0_INST);    

    while (1) {
        if (perform) {
            perform = false;
            DL_GPIO_setPins(GPIO_GRP_LEDS_PORT,
                GPIO_GRP_LEDS_PIN_LED_GREEN_PIN);

            perform_adc();
            
            perform_pid();

            perform_pwm();

            DL_GPIO_clearPins(GPIO_GRP_LEDS_PORT,
                GPIO_GRP_LEDS_PIN_LED_GREEN_PIN);
        }
    }
}


void perform_pid() {
    /* Execute PID control in every TM0 interrupt. */
    i32_Output_PID = PID(&PID_Var, i32_Target_Command, gAdcResult);
}

If you compare the perform_pid() with the one from the previous post, you 'll see that it now uses ADC as feedback, and output to drive the PWM.

Demo circuit

I used the same (random) low pass filter as in  MSP432 and TI-RTOS: PID Library Part 2 - Real World Example , to turn the PWM into a DC signal.

image

The filter sits between PWM out  (P26) and ADC in (PA25).

ccs project for EasyL1105: pid_EasyL1105_20251002_02.zip

Related posts

Parents

  • After a lot of tweaking, it's still not perfect, but I think acceptable.

    For the results in the chart below, I initially set a temperature of 70 degrees, and experimented with adding an aluminium thermal mass on top briefly to force the temperature down by a few degrees, and then removing it to see the response back to the set temperature.  I kept the mass on top at one stage (labeled Kept Mass), to see the temperature recover with the mass present too.

    Then, I reduced the set temperature by two degrees to 68 degC, then increased it back to 70 degrees, then set it to 150 degC, then again did the thermal mass test, then finally turned the set temperature down to 100 degC.

    From the chart, what can be determined is that there is overshoot, I struggled to tame that, and didn't succeed. The overshoot is usually under 1.5 degC, but under certain conditions (such as deliberately cooling the system by about 6-8 deg C by temporarily adding a thermal mass and then removing it), when it tries to recover, the overshoot can reach about 2.5 degC.

    The damping occurs reasonably well, the error is under 0.5 degC often within the first oscillation, before it dampens out. Maybe it would be better for the system to slowly approach the temperature so that it doesn't overshoot or undershoot at all.Or maybe I'm being too picky. Apparently 3D printer hot end control systems overshoot too, and even 5 degC is considered acceptable for that use-case.

    The undershoot shown in the chart below (when lowering the set temperature) is better than the overshoot; usually under 1 degC.

    The PID code is complicated now, it kicks in the 'I' portion only within a certain range for instance. I relied heavily on Chat GPT to handle that.

    In summary, although it functions, I'm sure performance could be improved a lot, by control system experts. But, for the actual end application (heating punches for leather), and probably a whole load of other applications, I think the current performance may be acceptable (it took an entire day of tests and tweaks to get it to this state: (

    I hope the control system doesn't have any unexpected behavior under certain conditions, but I'll find out over time, as it gets used. I need to move on to the PCB, to get the project into at least a usable state.

    image

  • in reply to shabaz

    This is great stuff. I'm impressed at how productive you both are here (for a part-time personal project). It looks pretty good to me for many purposes.

    I'm no control system expert (I don't have the practical experience to give good advice), but looking from the perspective of circuit feedback, I think the main complication is all the integrators in the loop. The low pass filter on the PWM is an integrator (at cutoff [10Hz, or so] the phase will be 45 degrees and move higher as you go up towards the PWM repetition rate of about 30Hz), any thermal mass on the output is an integrator sitting within the control loop [probably the dominant low pole, forcing the phase to 90 anywhere up near the PWM rate], and you have sampling and processing which will add other delays, so adding any more with the PID controller is going to make for a very fragile situation with, if not actual oscillation, a very poor phase margin. Since the main job of the PID integrator is to deal with the proportional term not being able to drive all the way to the set point [so as to mop up the very last bit of error], the 'hack' the AI has given you is to disable the I in all the other situations where there is enough gain to drive really embarrassing oscillation. It might be possible to tune the PID better - essentially the ID is a form of lag-lead compensation, so if you were to roughly model the load and the filter in electrical terms and draw a Bode plot of the open loop, you could maybe then see what was needed get a decent gain and phase margin - though I'm not sure how the sampled nature of the control and any latency is dealt with, and I'm not convinced a single D term can counter three integrators, can it?

    I don't know if this would work, but an alternative approach might be to turn off the PID I term entirely and dither the low bits of the set-point input (would need good random numbers with no DC offset, obviously), which might be enough to get the thermal mass integrator to the right point. Might be added complication for no real advantage though.

    Of course, if what you've got is good enough, stick with that. But if you want to develop it further, I think you are going to have to roughly model the 'mechanical' part of the loop [the 'plant'] and include it in your considerations.

  • in reply to jc2048

    Shabaz and I both have a different integrator. My PWM to DC setup has that RC filter.

    In Shabaz' design, the PWM isn't filtered, but it switches a MOSFET. And that fet manages a heating element with a big aluminium mass around it. A lot of mass, and a mega-integrator (and radiator Blush).

Comment Children
No Data