  13 August 2011
Author: Giorgos Lazaridis
PID Theory

A PID controller example explained in simple words

After 2 pages of theory and mathematical hocus-pocus, there will be people that still are not able to write software for a PID controller. For those people, i have 2 good news: First is that i found the best documentation for making a PID controller with a PIC microcontroller. It is the Application Note 937 from Microchip (who else): Implementing a PID Controller Using a PIC18 MCU [Microchip AN937]

Second good news: A PID controller example explained in simple words

I understand that not all people feel comfortable with mathematics, but many would like to make a PID controller for number of reasons. The integral term is by far the hardest to decipher. Instead of just explaining the integral term, i will explain a complete (and simple) PID controller with an example. I bet that after this, everyone will be able to make one (as long as he know how to program a microcontroller of course).

So, it goes like this: We want to control the temperature of a water heater. The heater resistor is controlled by a TRIAC, and the microcontroller controls the gate of that TRIAC (it could operate as a dimmer for example). A temperature sensor is placed in the water and gives feedback to the microcontroller in degrees Celsius (oC). The system has these parameters which you must define as RAM variables, and also it is good to save them in an EEPROM memory because you do not want to define them every time the system restarts:

• SP - The Set-Point of the system, the temperature that we want to achieve
• P_Gain - The gain for the Proportional term
• I_Gain - The gain for the Integral term
• D_Gain - The gain for the Derivative term
• I_Time - Reset time, this is the Ti parameter that we saw on previous pages
• D_Time - Derivative time, this is the Td parameter that we saw on previous pages

• You will also need to define these additional RAM variables:

• P_Term - The Proportional term to be added to the output.
• I_Term - The Integral term to be added to the output.
• D_Term - The Derivative term to be added to the output.
• PV - The Process Variable, this is the temperature that the sensor reads.
• Err - This will hold the current error (SP-PV).
• Acc_Err - This is the accumulated error to help calculate the Integral term
• D_Err - This is the derivative error to help calculate the Derivative term
• Ti_Counter - A countdown counter for the Integral accumulation routine
• Td_Counter - A countdown counter for the Derivative error calculation
• P_Out - The output power calculated by the PID

• This is the simplest PID controller. The output of this system is described by this function (do not be scared if you know little about mathematics) The microcontroller will run an infinite loop, during which it will calculate the PID terms and control the output. I call this routine "MainLoop". During start-up, the microcontroller will run the "Initialization" routine. Here are the flow chart of the Initialization and the MainLoop routines: (Click to enlarge)  Initialization (Click to enlarge) MainLoop (Click to enlarge)

Finally, there are 3 more subroutines, Calculate_P_Term, Calculate_I_Term and Calculate_D_Term. Here are the flow charts for those:   Calculate_P_Term(Click to enlarge) Calculate_I_Term(Click to enlarge) Calculate_D_Term(Click to enlarge)

How it works then?

First of all, i want you to agree that, behind the mask of the hard math, there is only a simple program with additions and subtractions (ok, and some multiplications and divisions maybe). I assume that you do know how to read a flowchart and i will not go into details. Regarding the symbols that i use: An ellipse indicates the start and the name of the (sub)routine that follows, a rectangle has the parts of a routine that have to be translated directly into code, a parallelogram defines a call to another sub-routine and a diamond defines an IF case, with a YES or NO branch.

Looking into the MainLoop, you can see that the system reads the temperature and calculates the error in every loop. The error is calculated by subtracting the PV from the SP (Err=SP-PV). If for example we want the water to have 80oC temperature (SP) and the current temperature is 25oC (PV), the error (Err) is 80-25=55oC. Using this error, the program calculates the P_Term (Calculate_P_Term subroutine), simply by multiplying the error by the P_Gain (P_Term=Err*P_Gain). Until now, this is the typical operation of a simple P-controller.

Now to add the I term. Looking again into the MainLoop, we see that the Calculate_I_Term subroutine is NOT called in every loop. Instead, a counter is used (Ti_Counter). The initial value of this counter is taken from the I_Time system parameter (Ti). The counter is decreased on every loop. When the counter becomes zero, the Calculate_I_Term subroutine is called. So, this is where the Ti parameter is used. Unlike the P_Term, the I_Term is calculated only once every Ti loop cycles. To calculate the I_Term the system adds the current error (Err) to a variable named Acc_Err. This variable accumulates all past errors and therefore is called Accumulated Error. The I_Term is then calculated by multiplying this variable by the I_Gain (I_Term=Acc_Err*I_Gain). Now you understand the purpose of the Ti parameter. If the system did not have this parameter, it would accumulate the current errors constantly, and therefore the I_Term would become huge and it would overshadow the P_Term and D_Term.

Finally, the D_Term. Similar to the I_Term, the D_Term has a different countdown timer (Td_Counter), so the Calculate_D_Term subroutine is called once every D_Time (Td) cycles. To calculate the D_Term, the system must know the previous error. The previous error is saved in the D_Err variable. So, the system can calculate the error difference by subtracting the previous error from the current error (Err - D_Err). Finally, the D_Term is calculated by multiplying the error difference by the D_Gain (D_Term = (Err - D_Err) * D_Gain).

The output of the system is calculated in every loop, and it is the sum of the 3 terms:

P_Out = P_Term + I_Term + D_Term

ATTENTION

There are some problems that may come out from the following example. First of all, you have to decide if you will use 8-bit registers of longer for the variables. This is something that you need to decide before you start designing the system. You have to take into account the system's range and accuracy. For example, a temperature controller that is used for a system which measures temperatures from 0 to 100 degrees with 1oC accuracy can be implemented with 8-bit registers, But if the system measures from 0 to 600oC then the 8-bits are just not enough.

Moreover, extreme caution must be taken with the P_Out and Acc_Err registers. Both of them are subject to overflow when big numbers are added. If for example an 8-bit registered is selected for the Acc_Err variable, but an error of 100 degrees is added 3 times, then the register will overflow and the result will be wrong. Same can happen to the P_out register if the result of the P and D term addition is larger than the register can handle. Therefore, make sure that you put some limits to these registers and check them every time to see if they exceed these limits.

Comments

 Name  Email (shall not be published)  Website Notify me of new posts via emailWrite your comments below:BEFORE you post a comment:You are welcome to comment for corrections and suggestions on this page. But if you have questions please use the forum instead to post it. Thank you. At 4 July 2014, 18:21:11 user Joseppi wrote:   [reply @ Joseppi]@Saimul If there is no oscillation then maybe you don't need a PID controller. Does it get unstable if you increase Kp too much? Is there a steady-state error?At 13 May 2014, 8:40:26 user Daro_PL wrote:   [reply @ Daro_PL]Great work! I'm also trying build PID controller to control temperature of heater. Heater heat the metal plate. Till this time: Heater is about 500W, and i trying control temperature by the triac. Heater is less frequently ON, if temperature is closer to the temperature that we want to achieve (set to 100deg.C). In this algorithm heater is on in duty from 10% do 100%. The best result is when: - duty=100% for T<70deg.C - duty=80% for T=71 to 80deg.C - duty=60% for T=81 to 90deg.C - duty=50% for T=91 to 100deg.C - when T>=100deg.C then heater is OFF In algorithm above temperature oscillates in about /- 3deg.C And that way I would like to implement PID controller. So for example if sum from calculating PID will be lets say from -50 to 120 then for my duty from 0 to 100% i must convert PID to range 0-100 so: PID=-50 to 120 DUTY=PID 50 DUTY=0 to 170 DUTY=DUTY/1.7 Is that correct translation ?At 13 February 2014, 13:45:08 user Saimul wrote:   [reply @ Saimul]Hi...I am working with a development of PID controller for a water heater. The problem that I am facing right now in determining the value of Kp,Ki and Kd. If I want to use Ziegler-Nicholas method for tuning PID then I have to determine the critical gain Kc and oscillation period Pc. After I can calculate PID parameters accordingly. But the problem is there is no oscillation in my system. So what should I do now? Thanks for your help.At 6 December 2013, 6:51:13 user Giorgos Lazaridis wrote:   [reply @ Giorgos Lazaridis]@sierra It largely depends on your application. If you have an 8-bit then you have to limit it from 0 to 255, or if you need negative numbers you limit to 127. So you have to do the proper divisions when you read the input to fit these ranges. Which means that you have to know before hands the minimum and maximum values that the sensor will send youAt 13 November 2013, 16:53:09 user sierra wrote:   [reply @ sierra]Hi, Nice work, thank you. When you calculate the P_Out value it may be too large or too small (negative). In such a case you may have to limit the maximum P_Out value. How do you determine the minimum and maximum P_Out value. RegardsAt 11 June 2013, 6:26:22 user Giorgos Lazaridis wrote:   [reply @ Giorgos Lazaridis]@yorick Of courseAt 5 June 2013, 9:53:34 user yorick wrote:   [reply @ yorick]@Giorgos Lazaridis @Giorgos Lazaridis moring can I use pic16f877 to program my pid using flowcode4?At 3 June 2013, 7:45:28 user Giorgos Lazaridis wrote:   [reply @ Giorgos Lazaridis]@Pushkin2023 fot Ti and Td i'd say the faster the better, depending on the response of the system (you want very quick response). As for the D term, do not reset it to zero, it will reset itself in a couple of Td times...At 3 June 2013, 7:41:23 user Giorgos Lazaridis wrote:   [reply @ Giorgos Lazaridis]@yorick PIC microcontrollers can work as PID of course. I have some examples in my circuit area.At 30 May 2013, 19:51:31 user Pushkin2023 wrote:   [reply @ Pushkin2023]Very nice tutorial. I'm trying to make a pid controller for my rotary inverted pendulum. I'm doing the pid computing every 20ms (timer2 interrupt, sets a flag in main loop), how should be Ti and Td?(how many times greater then pid loop) Another thing: the d_term and i_term shouldn't be reseted to 0 when the error is 0 or very very close to 0 ?At 29 May 2013, 6:02:27 user yorick wrote:   [reply @ yorick]Morning Thank you very much I was wondering how to adjust the set point and can I program it using flowcode mplab from microchip? If not what ic and program should I use ?At 10 May 2013, 17:07:31 user Chelnok wrote:   [reply @ Chelnok]Interesting. Finally starting to understand PID. Thanks :)At 27 April 2013, 7:43:29 user abhimanyu wrote:   [reply @ abhimanyu]great explanation! Thank You very much!At 19 February 2013, 22:19:59 user alfadex wrote:   [reply @ alfadex]Hi,where could i find info about p.i.d and how i can adjust these values buuut in Greek language?At 29 December 2012, 6:26:28 user Vishal wrote:   [reply @ Vishal]Nice post could you please give some information regarding autotuning of PID algo using C Code in Microcontroller. How to implement Zieglers Oscillation method to find ultimate gain and Period in C Code using PIC16 MicrocontrollerAt 23 December 2012, 19:02:12 user Yiannis wrote:   [reply @ Yiannis]mpravo re leventi !At 27 September 2012, 0:44:44 user Peter Lorry wrote:   [reply @ Peter Lorry]Can you confirm that this is correct and you sure don't mean the D term, it doesn't read correct to me. "The P term is increased slowly to increase the system response, but it must not become unstable"At 27 August 2012, 15:36:53 user Giorgos Lazaridis wrote:   [reply @ Giorgos Lazaridis]@NOAK yep!At 27 August 2012, 7:51:42 user NOAK wrote:   [reply @ NOAK]@Giorgos Lazaridis But most of the ADC in PIC uControllers for example are 10 bit, so its enough for up to 1023 right?At 24 August 2012, 16:21:43 user Giorgos Lazaridis wrote:   [reply @ Giorgos Lazaridis]@NOAK an 8-bit number can measure from 0 to 255 with integer numbers. So, it fully covers the range 0 to 100. But it cannot measure from 0 to 600 with 1 degree accuracy... Maybe with 3 degrees accuracy (because 3*256=768)At 24 August 2012, 11:12:34 user NOAK wrote:   [reply @ NOAK]"For example, a temperature controller that is used for a system which measures temperatures from 0 to 100 degrees with 1oC accuracy can be implemented with 8-bit registers, But if the system measures from 0 to 600oC then the 8-bits are just not enough." could u explain why 8-bit MCU is not enough for 0-600 C application?At 31 July 2012, 14:14:06 user akshay wrote:   [reply @ akshay]for example , i want to control a servo motor to move it to a commanded position and the output of the control system is PWM . so i should select proper P,I & D values so that the summation , i.e PID , is the correct PWM output which makes the motor move to the commanded position regardless to what degree the PWM should be propotional to the PID vlaue .At 29 July 2012, 20:28:51 user Giorgos Lazaridis wrote:   [reply @ Giorgos Lazaridis]@akshay That is the idea. You make all these calculations to come up with an output power that is required. This output can b anything, like an analog voltage, an angle of a servomotor, a throttle position of an engine.... anything.At 29 July 2012, 15:42:31 user akshay wrote:   [reply @ akshay]@Giorgos Lazaridis so the pwm output should be propotional to the pid value.At 26 July 2012, 17:58:13 user Giorgos Lazaridis wrote:   [reply @ Giorgos Lazaridis]@akshay now this depends on the control circuit. If it is PWM for example, you can change the duty cycle depending on the value you calculate... e.g. 0-100% duty cycle = 0 - 500 calculated PID Or if it has an analog output... you can change the analog value...At 26 July 2012, 14:36:19 user akshay wrote:   [reply @ akshay]I understood how the pid is calculated . but what do we do with the pid output value. how do we use the pid value to control the system.At 5 July 2012, 0:21:43 user Rob D wrote:   [reply @ Rob D]Great job explaining this! Now I get it. Thanks.At 10 March 2012, 21:20:03 user gyro_john wrote:   [reply @ gyro_john]Hi Giorgios, Thanks very much for the excellent info. I learned from you. I would like to give you and your readers links to another PID tutorial site which I liked very much. I hope that you might find it interesting: http://brettbeauregard.com/blog/2011/04/improving-the-beginners-pid-introduction/ ... it is 8 pages long! Filled with in-depth description and pictures the developmental changes to his code. And it is linked to the Arduino Playground code library, along with an Autotune library, at: http://arduino.cc/playground/Main/LibraryList in the Input / Output subcategory. Finally, the author links to the site of his own mentor, at: http://controlguru.com/ I have not yet looked at this myself, but I hope to find it educational. Once again thanks, and best regards.At 27 January 2012, 6:33:39 user Giorgos Lazaridis wrote:   [reply @ Giorgos Lazaridis]@Artur the idea of the It and Dt time parameters, is to avoid an overflow. The I parameter for example is used to have usually small changes, so that the operator can balance the immediate changes of the P parameter. If the I builds up on every cycle, then it will become huge very fast, and the system will not become stable. The idea is that the I build up slowly, and the system becomes stable slowly. The delay depends on the system's response. If a systems responses fast (for example hot air blower) then the It is small, otherwise (eg large water tank heater) It may be very large. Same stands for D. The parameter P follows the changes in output, it reads the error and changes the power accordingly in every cycle. The parameter I builds up slowly. Here is how we use to explain these parameters. P is the present, it looks at the output and changes with the present error. I is the past, it looks at the past errors and changes accordingly. D is the future, every Dt it looks at the error and according to the rate of change it tries to predict the future I am not sure what the result of the ring buffer would be. It sounds as a good idea, since the errors some 100 cycles before (for example) are not important for the I term. If i make another PID controller, i will experiment with this idea.At 26 January 2012, 21:34:05 user Artur wrote:   [reply @ Artur]I agree with others: best explanation I've seen so far. I'm wondering, would it be a good idea to store error in a ring buffer, e.g. 16 last values and calculate I and D from that on every iteration? If not then why. Other than the obvious added CPU cost.This way I would follow changes in output, not change incrementally every Ti cycles. Also why not calculate D in every loop based on previous loop's error?At 6 January 2012, 19:35:23 user Kammenos wrote:   [reply @ Kammenos]@Bill Kostelidis regarding the Integral, if you calculate it in every cycle, the system will never be stable, unless the system is highly reactive. That is exactly the point of the integral term existence after all: to check the progress between time intervals and decide if needs more or less power. To use the result of the PID, you need first to scale it to an easy-usable number. I use for example -127 to +127 which is one byte. The, you can either translate the number to 0-256 or just use it as is. Finally, you need to have a variable output. If the output is for example PWM, then i would translate the PID result to 0-256. Then, i would use this number as a PID duty cycle (0-0% 256=100%).At 4 January 2012, 23:48:07 user Bill Kostelidis wrote:   [reply @ Bill Kostelidis]Hello. I really like your projects. Although I have my disagreements with the way that the Integral is not calculated in every circle, since it is working, you must be correct! I am working in a PID project myself and I have a question, if you can help, it's great. I am heating water with immersion heaters, resistors. They are plugged directly in 220 Volts but I will use PWM to control them. The control is going to come from my PID. After I get the output from my PID controller, how am I going to translate it to PWM? I mean, if the output of my PID indicates 2 Celsius degrees down, how am I going to say to my PWM to adjust to 2 degrees down? I am going to work experimentally at first, but I am searching to find a more solid solution, any ideas? Thanks, Bill.At 27 October 2011, 4:01:13 user sorin wrote:   [reply @ sorin]Excellent info ! Thank you very much!At 10 October 2011, 7:51:00 user Le Van Chien wrote:   [reply @ Le Van Chien]I am a Vietnamese. I have read many textbooks about PID control in some subjects in university, but I've never really understood the meaning of this method, I think the theory is explained too complicatedly than what it really means. This post is a great explanation! I understood it easy. Thank you very much!At 30 September 2011, 16:20:54 user Erik T wrote:   [reply @ Erik T]This IS the best explanation of PID I have ever seen. The historical background adds so much to the explanation, that it makes so much more sense. I'd like to know how you created the graphs! I think a graphical demonstration program of PID would be awesome. It could allow users to set parameters and watch and compare settings. It could then show how to tune PID, using either method or both. I have retuned repaired PID controllers for CNC equipment, and until now have never understood why the directions for tuning were what they were. "... then reduce these two levels, and then turn this third one up until the machine oscillates." It all makes sense now. thank you, I look forward to coding something like the demo program because I cannot find one. Cheers!At 1 September 2011, 3:40:23 user Paulo de Almeida wrote:   [reply @ Paulo de Almeida]Excellent info ! Very well explained and illustrated. Thanks for sharing. Best regards, PA HOT in heaven!  NEW in heaven! New Theory: AC electric motor working principle