PID controller

A propor tional–integral–derivative controller (PID controller or three-term controller )

is a cont rol loop mechanism employing feedback t hat is widely used in indust rial cont rol
syst ems and a variet y of ot her applicat ions requiring cont inuously modulat ed cont rol. A PID
cont roller cont inuously calculat es an error value as t he difference bet ween a desired
set point (SP) and a measured process variable (PV) and applies a correct ion based on
proport ional, int egral, and derivat ive t erms (denot ed P, I, and D respect ively), hence t he name.

In pract ical t erms, PID aut omat ically applies an accurat e and responsive correct ion t o a
cont rol funct ion. An everyday example is t he cruise cont rol on a car, where ascending a hill
would lower speed if const ant engine power were applied. The cont roller's PID algorit hm
rest ores t he measured speed t o t he desired speed wit h minimal delay and overshoot by
increasing t he power out put of t he engine in a cont rolled manner.

The first t heoret ical analysis and pract ical applicat ion of PID was in t he field of aut omat ic
st eering syst ems for ships, developed from t he early 1920s onwards. It was t hen used for
aut omat ic process cont rol in t he manufact uring indust ry, where it was widely implement ed in
at first pneumat ic and t hen elect ronic cont rollers. Today t he PID concept is used universally
in applicat ions requiring accurat e and opt imized aut omat ic cont rol.

Fundamental operation
A block diagram of a PID controller in a feedback loop. r(t) is the desired process value or setpoint (S P), and y(t) is the
measured process value (PV).

The dist inguishing feat ure of t he PID cont roller is t he abilit y t o use t he t hree control terms
of proport ional, int egral and derivat ive influence on t he cont roller out put t o apply accurat e
and opt imal cont rol. The block diagram on t he right shows t he principles of how t hese t erms
are generat ed and applied. It shows a PID cont roller, which cont inuously calculat es an error
value as t he difference bet ween a desired set point and a measured process
variable : , and applies a correct ion based on proport ional,
int egral, and derivat ive t erms. The cont roller at t empt s t o minimize t he error over t ime by
adjust ment of a control variable , such as t he opening of a cont rol valve, t o a new value
det ermined by a weight ed sum of t he cont rol t erms.

In t his model:

Term P is proport ional t o t he current value of t he SP − PV error . For example, if t he

error is large, t he cont rol out put will be proport ionat ely large by using t he gain fact or "Kp".
Using proport ional cont rol alone will result in an error bet ween t he set point and t he
process value because t he cont roller requires an error t o generat e t he proport ional out put
response. In st eady st at e process condit ions an equilibrium is reached, wit h a st eady SP-
PV "offset ".

Term I account s for past values of t he SP − PV error and int egrat es t hem over t ime t o
produce t he I t erm. For example, if t here is a residual SP − PV error aft er t he applicat ion of
proport ional cont rol, t he int egral t erm seeks t o eliminat e t he residual error by adding a
cont rol effect due t o t he hist oric cumulat ive value of t he error. When t he error is
eliminat ed, t he int egral t erm will cease t o grow. This will result in t he proport ional effect
diminishing as t he error decreases, but t his is compensat ed for by t he growing int egral
effect .

Term D is a best est imat e of t he fut ure t rend of t he SP − PV error, based on it s current
rat e of change. It is somet imes called "ant icipat ory cont rol", as it is effect ively seeking t o
reduce t he effect of t he SP − PV error by exert ing a cont rol influence generat ed by t he
 rat e of error change. The more rapid t he change, t he great er t he cont rolling or damping
effect .[1]

Tuning – The balance of t hese effect s is achieved by loop t uning t o produce t he opt imal
cont rol funct ion. The t uning const ant s are shown below as "K" and must be derived for each
cont rol applicat ion, as t hey depend on t he response charact erist ics of t he complet e loop
ext ernal t o t he cont roller. These are dependent on t he behavior of t he measuring sensor, t he
final cont rol element (such as a cont rol valve), any cont rol signal delays and t he process
it self. Approximat e values of const ant s can usually be init ially ent ered knowing t he t ype of
applicat ion, but t hey are normally refined, or t uned, by "bumping" t he process in pract ice by
int roducing a set point change and observing t he syst em response.

Control action – The mat hemat ical model and pract ical loop above bot h use a direct cont rol
act ion for all t he t erms, which means an increasing posit ive error result s in an increasing
posit ive cont rol out put correct ion. The syst em is called reverse act ing if it is necessary t o
apply negat ive correct ive act ion. For inst ance, if t he valve in t he flow loop was 100–0% valve
opening for 0–100% cont rol out put – meaning t hat t he cont roller act ion has t o be reversed.
Some process cont rol schemes and final cont rol element s require t his reverse act ion. An
example would be a valve for cooling wat er, where t he fail-safe mode, in t he case of loss of
signal, would be 100% opening of t he valve; t herefore 0% cont roller out put needs t o cause
100% valve opening.

Mathematical form

The overall cont rol funct ion

where , , and , all non-negat ive, denot e t he coefficient s for t he proport ional,
int egral, and derivat ive t erms respect ively (somet imes denot ed P, I, and D).

In t he standard form of t he equat ion (see lat er in art icle), and are respect ively
replaced by and ; t he advant age of t his being t hat and have some
underst andable physical meaning, as t hey represent an int egrat ion t ime and a derivat ive t ime
respect ively. is t he t ime const ant wit h which t he cont roller will at t empt t o approach
t he set point . det ermines how long t he cont roller will t olerat e t he out put being
consist ent ly above or below t he set point .

Selective use of control terms

Alt hough a PID cont roller has t hree cont rol t erms, some applicat ions need only one or t wo
t erms t o provide appropriat e cont rol. This is achieved by set t ing t he unused paramet ers t o
zero and is called a PI, PD, P or I cont roller in t he absence of t he ot her cont rol act ions. PI
cont rollers are fairly common in applicat ions where derivat ive act ion would be sensit ive t o
measurement noise, but t he int egral t erm is oft en needed for t he syst em t o reach it s t arget
value.

Applicability

The use of t he PID algorit hm does not guarant ee opt imal cont rol of t he syst em or it s cont rol
st abilit y . Sit uat ions may occur where t here are excessive delays: t he measurement of t he
process value is delayed, or t he cont rol act ion does not apply quickly enough. In t hese cases
lead–lag compensat ion is required t o be effect ive. The response of t he cont roller can be
described in t erms of it s responsiveness t o an error, t he degree t o which t he syst em
overshoot s a set point , and t he degree of any syst em oscillat ion. But t he PID cont roller is
broadly applicable since it relies only on t he response of t he measured process variable, not
on knowledge or a model of t he underlying process.

History

Early PID theory was developed by observing the actions of helmsmen in keeping a vessel on course in the face of
varying influences such as wind and sea state.
Pneumatic PID (three-term) controller. The magnitudes of the three terms (P, I and D) are adjusted by the dials at the
top.

Origins

Cont inuous cont rol, before PID cont rollers were fully underst ood and implement ed, has one
of it s origins in t he cent rifugal governor, which uses rot at ing weight s t o cont rol a process.
This was invent ed by Christ iaan Huygens in t he 17t h cent ury t o regulat e t he gap bet ween
millst ones in windmills depending on t he speed of rot at ion, and t hereby compensat e for t he
variable speed of grain feed.[2][3]

Wit h t he invent ion of t he low-pressure st at ionary st eam engine t here was a need for
aut omat ic speed cont rol, and James Wat t ’s self-designed "conical pendulum" governor, a set
of revolving st eel balls at t ached t o a vert ical spindle by link arms, came t o be an indust ry
st andard. This was based on t he millst one-gap cont rol concept .[4]

Rot at ing-governor speed cont rol, however, was st ill variable under condit ions of varying load,
where t he short coming of what is now known as proport ional cont rol alone was evident . The
error bet ween t he desired speed and t he act ual speed would increase wit h increasing load. In
t he 19t h cent ury, t he t heoret ical basis for t he operat ion of governors was first described by
James Clerk Maxwell in 1868 in his now-famous paper On Governors . He explored t he
mat hemat ical basis for cont rol st abilit y, and progressed a good way t owards a solut ion, but
made an appeal for mat hemat icians t o examine t he problem.[5][4] The problem was examined
furt her in 1874 by Edward Rout h, Charles St urm, and in 1895, Adolf Hurwit z, all of whom
cont ribut ed t o t he est ablishment of cont rol st abilit y crit eria.[4] In subsequent applicat ions,
speed governors were furt her refined, not ably by American scient ist Willard Gibbs, who in
1872 t heoret ically analyzed Wat t 's conical pendulum governor.
About t his t ime, t he invent ion of t he Whit ehead t orpedo posed a cont rol problem t hat
required accurat e cont rol of t he running dept h. Use of a dept h pressure sensor alone proved
inadequat e, and a pendulum t hat measured t he fore and aft pit ch of t he t orpedo was
combined wit h dept h measurement t o become t he pendulum-and-hydrost at cont rol.
Pressure cont rol provided only a proport ional cont rol t hat , if t he cont rol gain was t oo high,
would become unst able and go int o overshoot wit h considerable inst abilit y of dept h-holding.
The pendulum added what is now known as derivat ive cont rol, which damped t he oscillat ions
by det ect ing t he t orpedo dive/climb angle and t hereby t he rat e-of-change of dept h.[6] This
development (named by Whit ehead as "The Secret " t o give no clue t o it s act ion) was around
1868.[7]

Anot her early example of a PID-t ype cont roller was developed by Elmer Sperry in 1911 for
ship st eering, t hough his work was int uit ive rat her t han mat hemat ically-based.[8]

It was not unt il 1922, however, t hat a formal cont rol law for what we now call PID or t hree-
t erm cont rol was first developed using t heoret ical analysis, by Russian American engineer
Nicolas Minorsky.[9] Minorsky was researching and designing aut omat ic ship st eering for t he
US Navy and based his analysis on observat ions of a helmsman. He not ed t he helmsman
st eered t he ship based not only on t he current course error but also on past error, as well as
t he current rat e of change;[10] t his was t hen given a mat hemat ical t reat ment by Minorsky.[4]
His goal was st abilit y, not general cont rol, which simplified t he problem significant ly. While
proport ional cont rol provided st abilit y against small dist urbances, it was insufficient for
dealing wit h a st eady dist urbance, not ably a st iff gale (due t o st eady-st at e error), which
required adding t he int egral t erm. Finally, t he derivat ive t erm was added t o improve st abilit y
and cont rol.

Trials were carried out on t he USS New Mexico, wit h t he cont rollers cont rolling t he angular
velocity (not t he angle) of t he rudder. PI cont rol yielded sust ained yaw (angular error) of ±2°.
Adding t he D element yielded a yaw error of ±1/6°, bet t er t han most helmsmen could
achieve.[11]

The Navy ult imat ely did not adopt t he syst em due t o resist ance by personnel. Similar work
was carried out and published by several ot hers in t he 1930s.

Industrial control
Proportional control using nozzle and flapper high gain amplifier and negative feedback

The wide use of feedback cont rollers did not become feasible unt il t he development of
wideband high-gain amplifiers t o use t he concept of negat ive feedback. This had been
developed in t elephone engineering elect ronics by Harold Black in t he lat e 1920s, but not
published unt il 1934.[4] Independent ly, Clesson E Mason of t he Foxboro Company in 1930
invent ed a wide-band pneumat ic cont roller by combining t he nozzle and flapper high-gain
pneumat ic amplifier, which had been invent ed in 1914, wit h negat ive feedback from t he
cont roller out put . This dramat ically increased t he linear range of operat ion of t he nozzle and
flapper amplifier, and int egral cont rol could also be added by t he use of a precision bleed
valve and a bellows generat ing t he int egral t erm. The result was t he "St abilog" cont roller
which gave bot h proport ional and int egral funct ions using feedback bellows.[4] The int egral
t erm was called Reset.[12] Lat er t he derivat ive t erm was added by a furt her bellows and
adjust able orifice.

From about 1932 onwards, t he use of wideband pneumat ic cont rollers increased rapidly in a
variet y of cont rol applicat ions. Air pressure was used for generat ing t he cont roller out put , and
also for powering process modulat ing devices such as diaphragm-operat ed cont rol valves.
They were simple low maint enance devices t hat operat ed well in harsh indust rial
environment s and did not present explosion risks in hazardous locat ions. They were t he
indust ry st andard for many decades unt il t he advent of discret e elect ronic cont rollers and
dist ribut ed cont rol syst ems (DCSs).

Wit h t hese cont rollers, a pneumat ic indust ry signaling st andard of 3–15 psi (0.2–1.0 bar) was
est ablished, which had an elevat ed zero t o ensure devices were working wit hin t heir linear
charact erist ic and represent ed t he cont rol range of 0-100%.

In t he 1950s, when high gain elect ronic amplifiers became cheap and reliable, elect ronic PID
cont rollers became popular, and t he pneumat ic st andard was emulat ed by 10-50 mA and 4–
20 mA current loop signals (t he lat t er became t he indust ry st andard). Pneumat ic field
act uat ors are st ill widely used because of t he advant ages of pneumat ic energy for cont rol
valves in process plant environment s.

S howing the evolution of analog control loop signaling from the pneumatic to the electronic eras

Current loops used for sensing and control signals. A modern electronic "smart" valve positioner is shown, which will
incorporate its own PID controller.

Most modern PID cont rols in indust ry are implement ed as comput er soft ware in DCSs,
programmable logic cont rollers (PLCs), or discret e compact cont rollers.

Electronic analog controllers

Elect ronic analog PID cont rol loops were oft en found wit hin more complex elect ronic
syst ems, for example, t he head posit ioning of a disk drive, t he power condit ioning of a power
supply, or even t he movement -det ect ion circuit of a modern seismomet er. Discret e
elect ronic analog cont rollers have been largely replaced by digit al cont rollers using
microcont rollers or FPGAs t o implement PID algorit hms. However, discret e analog PID
cont rollers are st ill used in niche applicat ions requiring high-bandwidt h and low-noise
performance, such as laser-diode cont rollers.[13]

Control loop example

Consider a robot ic arm[14] t hat can be moved and posit ioned by a cont rol loop. An elect ric
mot or may lift or lower t he arm, depending on forward or reverse power applied, but power
cannot be a simple funct ion of posit ion because of t he inert ial mass of t he arm, forces due
t o gravit y, ext ernal forces on t he arm such as a load t o lift or work t o be done on an ext ernal
object .

The sensed posit ion is t he process variable (PV).

The desired posit ion is called t he set point (SP).

The difference bet ween t he PV and SP is t he error (e), which quant ifies whet her t he arm is
t oo low or t oo high and by how much.

The input t o t he process (t he elect ric current in t he mot or) is t he out put from t he PID
cont roller. It is called eit her t he manipulat ed variable (MV) or t he cont rol variable (CV).

By measuring t he posit ion (PV), and subt ract ing it from t he set point (SP), t he error (e) is
found, and from it t he cont roller calculat es how much elect ric current t o supply t o t he mot or
(MV).

Proportional

The obvious met hod is propor tional cont rol: t he mot or current is set in proport ion t o t he
exist ing error. However, t his met hod fails if, for inst ance, t he arm has t o lift different weight s:
a great er weight needs a great er force applied for t he same error on t he down side, but a
smaller force if t he error is low on t he upside. That 's where t he int egral and derivat ive t erms
play t heir part .

Integral

An integral t erm increases act ion in relat ion not only t o t he error but also t he t ime for which
it has persist ed. So, if t he applied force is not enough t o bring t he error t o zero, t his force will
be increased as t ime passes. A pure "I" cont roller could bring t he error t o zero, but it would be
bot h slow react ing at t he st art (because t he act ion would be small at t he beginning,
depending on t ime t o get significant ) and brut al at t he end (t he act ion increases as long as
t he error is posit ive, even if t he error has st art ed t o approach zero).
Applying t oo much int egral when t he error is small and decreasing will lead t o overshoot . Aft er
overshoot ing, if t he cont roller were t o apply a large correct ion in t he opposit e direct ion and
repeat edly overshoot t he desired posit ion, t he out put would oscillat e around t he set point in
eit her a const ant , growing, or decaying sinusoid. If t he amplit ude of t he oscillat ions increases
wit h t ime, t he syst em is unst able. If t hey decrease, t he syst em is st able. If t he oscillat ions
remain at a const ant magnit ude, t he syst em is marginally st able.

Derivative

A derivative t erm does not consider t he magnit ude of t he error (meaning it cannot bring it t o
zero: a pure D cont roller cannot bring t he syst em t o it s set point ), but t he rat e of change of
error, t rying t o bring t his rat e t o zero. It aims at flat t ening t he error t raject ory int o a horizont al
line, damping t he force applied, and so reduces overshoot (error on t he ot her side because of
t oo great applied force).

Control damping

In t he int erest of achieving a cont rolled arrival at t he desired posit ion (SP) in a t imely and
accurat e way, t he cont rolled syst em needs t o be crit ically damped. A well-t uned posit ion
cont rol syst em will also apply t he necessary current s t o t he cont rolled mot or so t hat t he
arm pushes and pulls as necessary t o resist ext ernal forces t rying t o move it away from t he
required posit ion. The set point it self may be generat ed by an ext ernal syst em, such as a PLC
or ot her comput er syst em, so t hat it cont inuously varies depending on t he work t hat t he
robot ic arm is expect ed t o do. A well-t uned PID cont rol syst em will enable t he arm t o meet
t hese changing requirement s t o t he best of it s capabilit ies.

Response to disturbances

If a cont roller st art s from a st able st at e wit h zero error (PV = SP), t hen furt her changes by
t he cont roller will be in response t o changes in ot her measured or unmeasured input s t o t he
process t hat affect t he process, and hence t he PV. Variables t hat affect t he process ot her
t han t he MV are known as dist urbances. Generally, cont rollers are used t o reject dist urbances
and t o implement set point changes. A change in load on t he arm const it ut es a dist urbance t o
t he robot arm cont rol process.

Applications
In t heory, a cont roller can be used t o cont rol any process t hat has a measurable out put (PV),
a known ideal value for t hat out put (SP), and an input t o t he process (MV) t hat will affect t he
relevant PV. Cont rollers are used in indust ry t o regulat e t emperat ure, pressure, force, feed
rat e,[15] flow rat e, chemical composit ion (component concent rat ions), weight , posit ion, speed,
and pract ically every ot her variable for which a measurement exist s.

Controller theory

This section describes the parallel or non-interacting form of the PID controller. For other
forms please see the section Alternative nomenclature and forms.

The PID cont rol scheme is named aft er it s t hree correct ing t erms, whose sum const it ut es
t he manipulat ed variable (MV). The proport ional, int egral, and derivat ive t erms are summed t o
calculat e t he out put of t he PID cont roller. Defining as t he cont roller out put , t he final
form of t he PID algorit hm is

where

is t he proport ional gain, a t uning paramet er,

is t he int egral gain, a t uning paramet er,
is t he derivat ive gain, a t uning paramet er,
is t he error (SP is t he set point , and PV(t) is t he process variable),
is t he t ime or inst ant aneous t ime (t he present ),
is t he variable of int egrat ion (t akes on values from t ime 0 t o t he present ).

Equivalent ly, t he t ransfer funct ion in t he Laplace domain of t he PID cont roller is

where is t he complex frequency.

Proportional term
Response of PV to step change of S P vs time, for three values of Kp (Ki and Kd held constant)

The proport ional t erm produces an out put value t hat is proport ional t o t he current error value.
The proport ional response can be adjust ed by mult iplying t he error by a const ant Kp, called
t he proport ional gain const ant .

The proport ional t erm is given by

A high proport ional gain result s in a large change in t he out put for a given change in t he error.
If t he proport ional gain is t oo high, t he syst em can become unst able (see t he sect ion on loop
t uning). In cont rast , a small gain result s in a small out put response t o a large input error, and a
less responsive or less sensit ive cont roller. If t he proport ional gain is t oo low, t he cont rol
act ion may be t oo small when responding t o syst em dist urbances. Tuning t heory and
indust rial pract ice indicat e t hat t he proport ional t erm should cont ribut e t he bulk of t he
out put change.

Steady-state error

The steady-state error is t he difference bet ween t he desired final out put and t he act ual
one.[16] Because a non-zero error is required t o drive it , a proport ional cont roller generally
operat es wit h a st eady-st at e error.[a] St eady-st at e error (SSE) is proport ional t o t he process
gain and inversely proport ional t o proport ional gain. SSE may be mit igat ed by adding a
compensat ing bias t erm t o t he set point AND out put or correct ed dynamically by adding an
int egral t erm.

Integral term
Response of PV to step change of S P vs time, for three values of Ki (Kp and Kd held constant)

The cont ribut ion from t he int egral t erm is proport ional t o bot h t he magnit ude of t he error and
t he durat ion of t he error. The int egral in a PID cont roller is t he sum of t he inst ant aneous error
over t ime and gives t he accumulat ed offset t hat should have been correct ed previously. The
accumulat ed error is t hen mult iplied by t he int egral gain (Ki) and added t o t he cont roller
out put .

The int egral t erm is given by

The int egral t erm accelerat es t he movement of t he process t owards set point and eliminat es
t he residual st eady-st at e error t hat occurs wit h a pure proport ional cont roller. However, since
t he int egral t erm responds t o accumulat ed errors from t he past , it can cause t he present
value t o overshoot t he set point value (see t he sect ion on loop t uning).

Derivative term
Response of PV to step change of S P vs time, for three values of Kd (Kp and Ki held constant)

The derivat ive of t he process error is calculat ed by det ermining t he slope of t he error over
t ime and mult iplying t his rat e of change by t he derivat ive gain Kd. The magnit ude of t he
cont ribut ion of t he derivat ive t erm t o t he overall cont rol act ion is t ermed t he derivat ive gain,
Kd.

The derivat ive t erm is given by

Derivat ive act ion predict s syst em behavior and t hus improves set t ling t ime and st abilit y of
t he syst em.[17][18] An ideal derivat ive is not causal, so t hat implement at ions of PID cont rollers
include an addit ional low-pass filt ering for t he derivat ive t erm t o limit t he high-frequency gain
and noise. Derivat ive act ion is seldom used in pract ice t hough – by one est imat e in only 25%
of deployed cont rollers – because of it s variable impact on syst em st abilit y in real-world
applicat ions.

Loop tuning

Tuning a cont rol loop is t he adjust ment of it s cont rol paramet ers (proport ional band/gain,
int egral gain/reset , derivat ive gain/rat e) t o t he opt imum values for t he desired cont rol
response. St abilit y (no unbounded oscillat ion) is a basic requirement , but beyond t hat ,
different syst ems have different behavior, different applicat ions have different requirement s,
and requirement s may conflict wit h one anot her.
Even t hough t here are only t hree paramet ers and it is simple t o describe in principle, PID
t uning is a difficult problem because it must sat isfy complex crit eria wit hin t he limit at ions of
PID cont rol. Accordingly, t here are various met hods for loop t uning, and more sophist icat ed
t echniques are t he subject of pat ent s; t his sect ion describes some t radit ional, manual
met hods for loop t uning.

Designing and t uning a PID cont roller appears t o be concept ually int uit ive, but can be hard in
pract ice, if mult iple (and oft en conflict ing) object ives, such as short t ransient and high
st abilit y, are t o be achieved. PID cont rollers oft en provide accept able cont rol using default
t unings, but performance can generally be improved by careful t uning, and performance may
be unaccept able wit h poor t uning. Usually, init ial designs need t o be adjust ed repeat edly
t hrough comput er simulat ions unt il t he closed-loop syst em performs or compromises as
desired.

Some processes have a degree of nonlinearit y, so paramet ers t hat work well at full-load
condit ions do not work when t he process is st art ing up from no load. This can be correct ed
by gain scheduling (using different paramet ers in different operat ing regions).

Stability

If t he PID cont roller paramet ers (t he gains of t he proport ional, int egral and derivat ive t erms)
are chosen incorrect ly, t he cont rolled process input can be unst able; i.e., it s out put diverges,
wit h or wit hout oscillat ion, and is limit ed only by sat urat ion or mechanical breakage. Inst abilit y
is caused by excess gain, part icularly in t he presence of significant lag.

Generally, st abilizat ion of response is required and t he process must not oscillat e for any
combinat ion of process condit ions and set point s, t hough somet imes marginal st abilit y
(bounded oscillat ion) is accept able or desired.

Mat hemat ically, t he origins of inst abilit y can be seen in t he Laplace domain.[19]

The t ot al loop t ransfer funct ion is:

where is t he PID t ransfer funct ion and is t he plant t ransfer funct ion. A syst em is
unstable where t he closed loop t ransfer funct ion diverges for some .[19] This happens in
sit uat ions where . Typically, t his happens when wit h a
180-degree phase shift . St abilit y is guarant eed when for frequencies t hat
suffer high phase shift s. A more general formalism of t his effect is known as t he Nyquist
st abilit y crit erion.

Optimal behavior

The opt imal behavior on a process change or set point change varies depending on t he
applicat ion.

Two basic requirement s are regulation (dist urbance reject ion – st aying at a given set point )
and command tracking (implement ing set point changes). These t erms refer t o how well t he
cont rolled variable t racks t he desired value. Specific crit eria for command t racking include
rise t ime and set t ling t ime. Some processes must not allow an overshoot of t he process
variable beyond t he set point if, for example, t his would be unsafe. Ot her processes must
minimize t he energy expended in reaching a new set point .

Overview of tuning methods

There are several met hods for t uning a PID loop. The most effect ive met hods generally
involve developing some form of process model and t hen choosing P, I, and D based on t he
dynamic model paramet ers. Manual t uning met hods can be relat ively t ime-consuming,
part icularly for syst ems wit h long loop t imes.

The choice of met hod depends largely on whet her t he loop can be t aken offline for t uning,
and on t he response t ime of t he syst em. If t he syst em can be t aken offline, t he best t uning
met hod oft en involves subject ing t he syst em t o a st ep change in input , measuring t he out put
as a funct ion of t ime, and using t his response t o det ermine t he cont rol paramet ers.
Choosing a tuning method

Method Advantages Disadvantages

Requires
Manual
No mat hemat ics required; online. experienced
tuning
personnel.

Process upset ,
Ziegler– some t rial-and-
Nichols Proven met hod; online. error, very
[b]
aggressive
t uning.

Process upset ,
some t rial-and-
Tyreus
Proven met hod; online. error, very
Luyben
aggressive
t uning.

Consist ent t uning; online or offline – can employ comput er-

Some cost or
Software aut omat ed cont rol syst em design (CAutoD) t echniques; may
t raining
tools include valve and sensor analysis; allows simulat ion before
involved.[21]
downloading; can support non-st eady-st at e (NSS) t uning.

Some
mat hemat ics;
Cohen– offline; only
Good process models.
Coon good for first -
order
processes.

The process
Åström- Can be used for aut o t uning; amplit ude is minimum so t his it self is
Hägglund met hod has lowest process upset inherent ly
oscillat ory.

Manual tuning

If t he syst em must remain online, one t uning met hod is t o first set and values t o zero.
Increase t he unt il t he out put of t he loop oscillat es; t hen set t o approximat ely half
t hat value for a "quart er amplit ude decay"-t ype response. Then increase unt il any offset
is correct ed in sufficient t ime for t he process, but not unt il t oo great a value causes
inst abilit y. Finally, increase , if required, unt il t he loop is accept ably quick t o reach it s
reference aft er a load dist urbance. Too much causes excessive response and overshoot .
A fast PID loop t uning usually overshoot s slight ly t o reach t he set point more quickly;
however, some syst ems cannot accept overshoot , in which case an overdamped closed-loop
syst em is required, which in t urn requires a set t ing significant ly less t han half t hat of t he
set t ing t hat was causing oscillat ion.

Effects of varying PID parameters (Kp,Ki,Kd ) on the step response of a system

Effects of increasing a parameter independently[22][23]

Settling Steady-state
Parameter Rise time Overshoot Stability
time error

Small
Decrease Increase Decrease Degrade
change

Decrease Increase Increase Eliminat e Degrade

Minor No effect in Improve if

Decrease Decrease
change t heory small

Ziegler–Nichols method

Anot her heurist ic t uning met hod is known as t he Ziegler–Nichols met hod, int roduced by John
G. Ziegler and Nat haniel B. Nichols in t he 1940s. As in t he met hod above, t he and gains
are first set t o zero. The proport ional gain is increased unt il it reaches t he ult imat e gain, ,
at which t he out put of t he loop st art s t o oscillat e const ant ly. and t he oscillat ion period
are used t o set t he gains as follows:

Ziegler–Nichols method

Control Type

P — —

PI —

PID

These gains apply t o t he ideal, parallel form of t he PID cont roller. When applied t o t he
st andard PID form, only t he int egral and derivat ive gains and are dependent on t he
oscillat ion period .

Cohen–Coon parameters

This met hod was developed in 1953 and is based on a first -order + t ime delay model. Similar
t o t he Ziegler–Nichols met hod, a set of t uning paramet ers were developed t o yield a closed-
loop response wit h a decay rat io of . Arguably t he biggest problem wit h t hese paramet ers is
t hat a small change in t he process paramet ers could pot ent ially cause a closed-loop syst em
t o become unst able.

Relay (Åström–Hägglund) method

Published in 1984 by Karl Johan Åst röm and Tore Hägglund,[24] t he relay met hod t emporarily
operat es t he process using bang-bang cont rol and measures t he result ant oscillat ions. The
out put is swit ched (as if by a relay, hence t he name) bet ween t wo values of t he cont rol
variable. The values must be chosen so t he process will cross t he set point , but t hey need not
be 0% and 100%; by choosing suit able values, dangerous oscillat ions can be avoided.

As long as t he process variable is below t he set point , t he cont rol out put is set t o t he higher
value. As soon as it rises above t he set point , t he cont rol out put is set t o t he lower value.
Ideally, t he out put waveform is nearly square, spending equal t ime above and below t he
set point . The period and amplit ude of t he result ant oscillat ions are measured, and used t o
comput e t he ult imat e gain and period, which are t hen fed int o t he Ziegler–Nichols met hod.

Specifically, t he ult imat e period is assumed t o be equal t o t he observed period, and t he

ult imat e gain is comput ed as where a is t he amplit ude of t he process variable
oscillat ion, and b is t he amplit ude of t he cont rol out put change which caused it .
There are numerous variant s on t he relay met hod.[25]

First order with dead time model

The t ransfer funct ion for a first -order process, wit h dead t ime, is:

where kp is t he process gain, τ p is t he t ime const ant , θ is t he dead t ime, and u(s) is a st ep
change input . Convert ing t his t ransfer funct ion t o t he t ime domain result s in:

using t he same paramet ers found above.

It is import ant when using t his met hod t o apply a large enough st ep change input t hat t he
out put can be measured; however, t oo large of a st ep change can affect t he process
st abilit y. Addit ionally, a larger st ep change ensures t hat t he out put does not change due t o a
dist urbance (for best result s, t ry t o minimize dist urbances when performing t he st ep t est ).

One way t o det ermine t he paramet ers for t he first -order process is using t he 63.2% met hod.
In t his met hod, t he process gain (kp) is equal t o t he change in out put divided by t he change in
input . The dead t ime (θ) is t he amount of t ime bet ween when t he st ep change occurred and
when t he out put first changed. The t ime const ant (τ p) is t he amount of t ime it t akes for t he
out put t o reach 63.2% of t he new st eady-st at e value aft er t he st ep change. One downside
t o using t his met hod is t hat it can t ake a while t o reach a new st eady-st at e value if t he
process has large t ime const ant s.[26]

Tuning software

Most modern indust rial facilit ies no longer t une loops using t he manual calculat ion met hods
shown above. Inst ead, PID t uning and loop opt imizat ion soft ware are used t o ensure
consist ent result s. These soft ware packages gat her dat a, develop process models, and
suggest opt imal t uning. Some soft ware packages can even develop t uning by gat hering dat a
from reference changes.

Mat hemat ical PID loop t uning induces an impulse in t he syst em and t hen uses t he cont rolled
syst em's frequency response t o design t he PID loop values. In loops wit h response t imes of
several minut es, mat hemat ical loop t uning is recommended, because t rial and error can t ake
days just t o find a st able set of loop values. Opt imal values are harder t o find. Some digit al
loop cont rollers offer a self-t uning feat ure in which very small set point changes are sent t o
t he process, allowing t he cont roller it self t o calculat e opt imal t uning values.

Anot her approach calculat es init ial values via t he Ziegler–Nichols met hod, and uses a
numerical opt imizat ion t echnique t o find bet t er PID coefficient s.[27]

Ot her formulas are available t o t une t he loop according t o different performance crit eria.
Many pat ent ed formulas are now embedded wit hin PID t uning soft ware and hardware
modules.[28]

Advances in aut omat ed PID loop t uning soft ware also deliver algorit hms for t uning PID Loops
in a dynamic or non-st eady st at e (NSS) scenario. The soft ware models t he dynamics of a
process, t hrough a dist urbance, and calculat e PID cont rol paramet ers in response.[29]

Limitations

While PID cont rollers are applicable t o many cont rol problems, and oft en perform
sat isfact orily wit hout any improvement s or only coarse t uning, t hey can perform poorly in
some applicat ions and do not in general provide optimal cont rol. The fundament al difficult y
wit h PID cont rol is t hat it is a feedback cont rol syst em, wit h constant paramet ers, and no
direct knowledge of t he process, and t hus overall performance is react ive and a compromise.
While PID cont rol is t he best cont roller in an observer wit hout a model of t he process, bet t er
performance can be obt ained by overt ly modeling t he act or of t he process wit hout resort ing
t o an observer.

PID cont rollers, when used alone, can give poor performance when t he PID loop gains must
be reduced so t hat t he cont rol syst em does not overshoot , oscillat e or hunt about t he
cont rol set point value. They also have difficult ies in t he presence of non-linearit ies, may
t rade-off regulat ion versus response t ime, do not react t o changing process behavior (say,
t he process changes aft er it has warmed up), and have lag in responding t o large
dist urbances.

The most significant improvement is t o incorporat e feed-forward cont rol wit h knowledge
about t he syst em, and using t he PID only t o cont rol error. Alt ernat ively, PIDs can be modified
in more minor ways, such as by changing t he paramet ers (eit her gain scheduling in different
use cases or adapt ively modifying t hem based on performance), improving measurement
(higher sampling rat e, precision, and accuracy, and low-pass filt ering if necessary), or
cascading mult iple PID cont rollers.
Linearity and symmetry

PID cont rollers work best when t he loop t o be cont rolled is linear and symmet ric. Thus, t heir
performance in non-linear and asymmet ric syst ems is degraded.

A non-linear valve, for inst ance, in a flow cont rol applicat ion, will result in variable loop
sensit ivit y, requiring dampened act ion t o prevent inst abilit y. One solut ion is t he use of t he
valve's non-linear charact erist ic in t he cont rol algorit hm t o compensat e for t his.

An asymmet ric applicat ion, for example, is t emperat ure cont rol in HVAC syst ems using only
act ive heat ing (via a heat ing element ), where t here is only passive cooling available. When it is
desired t o lower t he cont rolled t emperat ure t he heat ing out put is off, but t here is no act ive
cooling due t o cont rol out put . Any overshoot of rising t emperat ure can t herefore only be
correct ed slowly; it cannot be forced downward by t he cont rol out put . In t his case t he PID
cont roller could be t uned t o be over-damped, t o prevent or reduce overshoot , but t his
reduces performance by increasing t he set t ling t ime of a rising t emperat ure t o t he set point .
The inherent degradat ion of cont rol qualit y in t his applicat ion could be solved by applicat ion
of act ive cooling.

Noise in derivative term

A problem wit h t he derivat ive t erm is t hat it amplifies higher frequency measurement or
process noise t hat can cause large amount s of change in t he out put . It is oft en helpful t o
filt er t he measurement s wit h a low-pass filt er in order t o remove higher-frequency noise
component s. As low-pass filt ering and derivat ive cont rol can cancel each ot her out , t he
amount of filt ering is limit ed. Therefore, low noise inst rument at ion can be import ant . A
nonlinear median filt er may be used, which improves t he filt ering efficiency and pract ical
performance.[30] In some cases, t he different ial band can be t urned off wit h lit t le loss of
cont rol. This is equivalent t o using t he PID cont roller as a PI cont roller.

Modifications to the algorithm

The basic PID algorit hm present s some challenges in cont rol applicat ions t hat have been
addressed by minor modificat ions t o t he PID form.

Integral windup

One common problem result ing from t he ideal PID implement at ions is int egral windup.
Following a large change in set point t he int egral t erm can accumulat e an error larger t han t he
maximal value for t he regulat ion variable (windup), t hus t he syst em overshoot s and cont inues
t o increase unt il t his accumulat ed error is unwound. This problem can be addressed by:

Disabling t he int egrat ion unt il t he PV has ent ered t he cont rollable region

Prevent ing t he int egral t erm from accumulat ing above or below pre-det ermined bounds

Back-calculat ing t he int egral t erm t o const rain t he regulat or out put wit hin feasible
bounds.[31]

Overshooting from known disturbances

For example, a PID loop is used t o cont rol t he t emperat ure of an elect ric resist ance furnace
where t he syst em has st abilized. Now when t he door is opened and somet hing cold is put int o
t he furnace t he t emperat ure drops below t he set point . The int egral funct ion of t he cont roller
t ends t o compensat e for error by int roducing anot her error in t he posit ive direct ion. This
overshoot can be avoided by freezing of t he int egral funct ion aft er t he opening of t he door
for t he t ime t he cont rol loop t ypically needs t o reheat t he furnace.

PI controller

Basic block of a PI controller

A PI controller (proport ional-int egral cont roller) is a special case of t he PID cont roller in
which t he derivat ive (D) of t he error is not used.

The cont roller out put is given by

where is t he error or deviat ion of act ual measured value (PV) from t he set point (S P).
A PI cont roller can be modelled easily in soft ware such as Simulink or Xcos using a "flow
chart " box involving Laplace operat ors:

where

= proport ional gain

= int egral gain

Set t ing a value for is oft en a t rade off bet ween decreasing overshoot and increasing
set t ling t ime.

The lack of derivat ive act ion may make t he syst em more st eady in t he st eady st at e in t he
case of noisy dat a. This is because derivat ive act ion is more sensit ive t o higher-frequency
t erms in t he input s.

Wit hout derivat ive act ion, a PI-cont rolled syst em is less responsive t o real (non-noise) and
relat ively fast alt erat ions in st at e and so t he syst em will be slower t o reach set point and
slower t o respond t o pert urbat ions t han a well-t uned PID syst em may be.

Deadband

Many PID loops cont rol a mechanical device (for example, a valve). Mechanical maint enance
can be a major cost and wear leads t o cont rol degradat ion in t he form of eit her st ict ion or
backlash in t he mechanical response t o an input signal. The rat e of mechanical wear is mainly
a funct ion of how oft en a device is act ivat ed t o make a change. Where wear is a significant
concern, t he PID loop may have an out put deadband t o reduce t he frequency of act ivat ion of
t he out put (valve). This is accomplished by modifying t he cont roller t o hold it s out put st eady
if t he change would be small (wit hin t he defined deadband range). The calculat ed out put
must leave t he deadband before t he act ual out put will change.

Setpoint step change

The proport ional and derivat ive t erms can produce excessive movement in t he out put when a
syst em is subject ed t o an inst ant aneous st ep increase in t he error, such as a large set point
change. In t he case of t he derivat ive t erm, t his is due t o t aking t he derivat ive of t he error,
which is very large in t he case of an inst ant aneous st ep change. As a result , some PID
algorit hms incorporat e some of t he following modificat ions:
 Setpoint ramping
In t his modificat ion, t he set point is gradually moved from it s old value t o a newly specified
value using a linear or first -order different ial ramp funct ion. This avoids t he discont inuit y
present in a simple st ep change.
Derivative of the process variable
In t his case t he PID cont roller measures t he derivat ive of t he measured process variable
(PV), rat her t han t he derivat ive of t he error. This quant it y is always cont inuous (i.e., never
has a st ep change as a result of changed set point ). This modificat ion is a simple case of
set point weight ing.
Setpoint weighting
Set point weight ing adds adjust able fact ors (usually bet ween 0 and 1) t o t he set point in t he
error in t he proport ional and derivat ive element of t he cont roller. The error in t he int egral
t erm must be t he t rue cont rol error t o avoid st eady-st at e cont rol errors. These t wo ext ra
paramet ers do not affect t he response t o load dist urbances and measurement noise and
can be t uned t o improve t he cont roller's set point response.

Feed-forward

The cont rol syst em performance can be improved by combining t he feedback (or closed-
loop) cont rol of a PID cont roller wit h feed-forward (or open-loop) cont rol. Knowledge about
t he syst em (such as t he desired accelerat ion and inert ia) can be fed forward and combined
wit h t he PID out put t o improve t he overall syst em performance. The feed-forward value
alone can oft en provide t he major port ion of t he cont roller out put . The PID cont roller
primarily has t o compensat e for what ever difference or error remains bet ween t he set point
(SP) and t he syst em response t o t he open-loop cont rol. Since t he feed-forward out put is not
affect ed by t he process feedback, it can never cause t he cont rol syst em t o oscillat e, t hus
improving t he syst em response wit hout affect ing st abilit y. Feed forward can be based on t he
set point and on ext ra measured dist urbances. Set point weight ing is a simple form of feed
forward.

For example, in most mot ion cont rol syst ems, in order t o accelerat e a mechanical load under
cont rol, more force is required from t he act uat or. If a velocit y loop PID cont roller is being
used t o cont rol t he speed of t he load and command t he force being applied by t he act uat or,
t hen it is beneficial t o t ake t he desired inst ant aneous accelerat ion, scale t hat value
appropriat ely and add it t o t he out put of t he PID velocit y loop cont roller. This means t hat
whenever t he load is being accelerat ed or decelerat ed, a proport ional amount of force is
commanded from t he act uat or regardless of t he feedback value. The PID loop in t his
sit uat ion uses t he feedback informat ion t o change t he combined out put t o reduce t he
remaining difference bet ween t he process set point and t he feedback value. Working
t oget her, t he combined open-loop feed-forward cont roller and closed-loop PID cont roller
can provide a more responsive cont rol syst em.

Bumpless operation

PID cont rollers are oft en implement ed wit h a "bumpless" init ializat ion feat ure t hat
recalculat es t he int egral accumulat or t erm t o maint ain a consist ent process out put t hrough
paramet er changes.[32] A part ial implement at ion is t o st ore t he int egral gain t imes t he error
rat her t han st oring t he error and post mult iplying by t he int egral gain, which prevent s
discont inuous out put when t he I gain is changed, but not t he P or D gains.

Other improvements

In addit ion t o feed-forward, PID cont rollers are oft en enhanced t hrough met hods such as
PID gain scheduling (changing paramet ers in different operat ing condit ions), fuzzy logic, or
comput at ional verb logic.[33][34] Furt her pract ical applicat ion issues can arise from
inst rument at ion connect ed t o t he cont roller. A high enough sampling rat e, measurement
precision, and measurement accuracy are required t o achieve adequat e cont rol performance.
Anot her new met hod for improvement of PID cont roller is t o increase t he degree of freedom
by using fract ional order. The order of t he int egrat or and different iat or add increased
flexibilit y t o t he cont roller.[35]

Cascade control

One dist inct ive advant age of PID cont rollers is t hat t wo PID cont rollers can be used
t oget her t o yield bet t er dynamic performance. This is called cascaded PID cont rol. Two
cont rollers are in cascade when t hey are arranged so t hat one regulat es t he set point of t he
ot her. A PID cont roller act s as out er loop cont roller, which cont rols t he primary physical
paramet er, such as fluid level or velocit y. The ot her cont roller act s as inner loop cont roller,
which reads t he out put of out er loop cont roller as set point , usually cont rolling a more rapid
changing paramet er, flowrat e or accelerat ion. It can be mat hemat ically proven t hat t he
working frequency of t he cont roller is increased and t he t ime const ant of t he object is
reduced by using cascaded PID cont rollers..

For example, a t emperat ure-cont rolled circulat ing bat h has t wo PID cont rollers in cascade,
each wit h it s own t hermocouple t emperat ure sensor. The out er cont roller cont rols t he
t emperat ure of t he wat er using a t hermocouple locat ed far from t he heat er, where it
accurat ely reads t he t emperat ure of t he bulk of t he wat er. The error t erm of t his PID
cont roller is t he difference bet ween t he desired bat h t emperat ure and measured
t emperat ure. Inst ead of cont rolling t he heat er direct ly, t he out er PID cont roller set s a heat er
t emperat ure goal for t he inner PID cont roller. The inner PID cont roller cont rols t he
t emperat ure of t he heat er using a t hermocouple at t ached t o t he heat er. The inner
cont roller's error t erm is t he difference bet ween t his heat er t emperat ure set point and t he
measured t emperat ure of t he heat er. It s out put cont rols t he act ual heat er t o st ay near t his
set point .

The proport ional, int egral, and different ial t erms of t he t wo cont rollers will be very different .
The out er PID cont roller has a long t ime const ant – all t he wat er in t he t ank needs t o heat up
or cool down. The inner loop responds much more quickly. Each cont roller can be t uned t o
mat ch t he physics of t he syst em it cont rols – heat t ransfer and t hermal mass of t he whole
t ank or of just t he heat er – giving bet t er t ot al response.[36][37]

Alternative nomenclature and forms

Standard versus parallel (ideal) form

The form of t he PID cont roller most oft en encount ered in indust ry, and t he one most
relevant t o t uning algorit hms is t he standard form. In t his form t he gain is applied t o t he
, and t erms, yielding:

where

is t he integral time
is t he derivative time

In t his st andard form, t he paramet ers have a clear physical meaning. In part icular, t he inner
summat ion produces a new single error value which is compensat ed for fut ure and past
errors. The proport ional error t erm is t he current error. The derivat ive component s t erm
at t empt s t o predict t he error value at seconds (or samples) in t he fut ure, assuming t hat
t he loop cont rol remains unchanged. The int egral component adjust s t he error value t o
compensat e for t he sum of all past errors, wit h t he int ent ion of complet ely eliminat ing t hem
in seconds (or samples). The result ing compensat ed single error value is t hen scaled by
t he single gain t o comput e t he cont rol variable.

In t he parallel form, shown in t he cont roller t heory sect ion

t he gain paramet ers are relat ed t o t he paramet ers of t he st andard form t hrough
and . This parallel form, where t he paramet ers are t reat ed as
simple gains, is t he most general and flexible form. However, it is also t he form where t he
paramet ers have t he weakest relat ionship t o physical behaviors and is generally reserved for
t heoret ical t reat ment of t he PID cont roller. The st andard form, despit e being slight ly more
complex mat hemat ically, is more common in indust ry.

Reciprocal gain, a.k.a. proportional band

In many cases, t he manipulat ed variable out put by t he PID cont roller is a dimensionless
fract ion bet ween 0 and 100% of some maximum possible value, and t he t ranslat ion int o real
unit s (such as pumping rat e or wat t s of heat er power) is out side t he PID cont roller. The
process variable, however, is in dimensioned unit s such as t emperat ure. It is common in t his
case t o express t he gain not as "out put per degree", but rat her in t he reciprocal form of a
proportional band , which is "degrees per full out put ": t he range over which t he
out put changes from 0 t o 1 (0% t o 100%). Beyond t his range, t he out put is sat urat ed, full-off
or full-on. The narrower t his band, t he higher t he proport ional gain.

Basing derivative action on PV

In most commercial cont rol syst ems, derivat ive act ion is based on process variable rat her
t han error. That is, a change in t he set point does not affect t he derivat ive act ion. This is
because t he digit ized version of t he algorit hm produces a large unwant ed spike when t he
set point is changed. If t he set point is const ant t hen changes in t he PV will be t he same as
changes in error. Therefore, t his modificat ion makes no difference t o t he way t he cont roller
responds t o process dist urbances.

Basing proportional action on PV

Most commercial cont rol syst ems offer t he option of also basing t he proport ional act ion
solely on t he process variable. This means t hat only t he int egral act ion responds t o changes
in t he set point . The modificat ion t o t he algorit hm does not affect t he way t he cont roller
responds t o process dist urbances. Basing proport ional act ion on PV eliminat es t he inst ant
and possibly very large change in out put caused by a sudden change t o t he set point .
Depending on t he process and t uning t his may be beneficial t o t he response t o a set point
st ep.
King[38] describes an effect ive chart -based met hod.

Laplace form

Somet imes it is useful t o writ e t he PID regulat or in Laplace t ransform form:

Having t he PID cont roller writ t en in Laplace form and having t he t ransfer funct ion of t he
cont rolled syst em makes it easy t o det ermine t he closed-loop t ransfer funct ion of t he
syst em.

Series/interacting form

Anot her represent at ion of t he PID cont roller is t he series, or interacting form

where t he paramet ers are relat ed t o t he paramet ers of t he st andard form t hrough

, , and

wit h

This form essent ially consist s of a PD and PI cont roller in series. As t he int egral is required t o
calculat e t he cont roller's bias t his form provides t he abilit y t o t rack an ext ernal bias value
which is required t o be used for proper implement at ion of mult i-cont roller advanced cont rol
schemes.

Discrete implementation

The analysis for designing a digit al implement at ion of a PID cont roller in a microcont roller
(MCU) or FPGA device requires t he st andard form of t he PID cont roller t o be discretized.[39]
Approximat ions for first -order derivat ives are made by backward finit e differences. and
are discret ized wit h a sampling period , k is t he sample index.

Different iat ing bot h sides of PID equat ion using Newt on's not at ion gives:

Derivat ive t erms are approximat ed as,

So,

Applying backward difference again gives,

By simplifying and regrouping t erms of t he above equat ion, an algorit hm for an

implement at ion of t he discret ized PID cont roller in a MCU is finally obt ained:

or:

s.t .

Not e: This met hod solves in fact

where is a const ant independent of t . This const ant is useful when you want t o have a
st art and st op cont rol on t he regulat ion loop. For inst ance, set t ing Kp,Ki and Kd t o 0 will keep
u(t ) const ant . Likewise, when you want t o st art a regulat ion on a syst em where t he error is
already close t o 0 wit h u(t ) non null, it prevent s from sending t he out put t o 0.

Pseudocode
Here is a very simple and explicit group of pseudocode t hat can be easily underst ood by t he
layman:

Kp - proport ional gain

Ki - int egral gain

Kd - derivat ive gain

dt - loop int erval t ime

previous_error := 0
integral := 0
loop:
error := setpoint − measured_value
proportional := error;
integral := integral + error × dt
derivative := (error − previous_error) / dt
output := Kp × proportional + Ki × integral + Kd ×
derivative
previous_error := error
wait(dt)
goto loop

Here is a more complicat ed and much less explicit soft ware loop t hat implement s a PID
algorit hm:

A0 := Kp + Ki*dt + Kd/dt
A1 := -Kp - 2*Kd/dt
A2 := Kd/dt
error[2] := 0 // e(t-2)
error[1] := 0 // e(t-1)
error[0] := 0 // e(t)
output := u0 // Usually the current value of the actuator

loop:
error[2] := error[1]
error[1] := error[0]
error[0] := setpoint − measured_value
output := output + A0 * error[0] + A1 * error[1] + A2 *
error[2]
 wait(dt)
goto loop

Here, Kp is a dimensionless number, Ki is expressed in and Kd is expressed in s. When

doing a regulat ion where t he act uat or and t he measured value are not in t he same unit (ex.
t emperat ure regulat ion using a mot or cont rolling a valve), Kp, Ki and Kd may be correct ed by a
unit conversion fact or. It may also be int erest ing t o use Ki in it s reciprocal form (int egrat ion
t ime). The above implement at ion allows t o perform a I-only cont roller which may be useful in
some cases.

In t he real world, t his is D-t o-A convert ed and passed int o t he process under cont rol as t he
manipulat ed variable (MV). The current error is st ored elsewhere for re-use in t he next
different iat ion, t he program t hen wait s unt il dt seconds have passed since st art , and t he loop
begins again, reading in new values for t he PV and t he set point and calculat ing a new value for
t he error.[40]

Not e t hat for real code, t he use of "wait (dt )" might be inappropriat e because it doesn't
account for t ime t aken by t he algorit hm it self during t he loop, or more import ant ly, any
preempt ion delaying t he algorit hm.

A common issue when using is t he response t o t he derivat ive of a rising or falling edge of
t he set point as shown below:
A t ypical workaround is t o filt er t he derivat ive act ion using a low pass filt er of t ime const ant
where :
A variant of t he above algorit hm using an Infinit e impulse response filt er for t he derivat ive:

A0 := Kp + Ki*dt
A1 := -Kp
error[2] := 0 // e(t-2)
error[1] := 0 // e(t-1)
error[0] := 0 // e(t)
output := u0 // Usually the current value of the actuator
A0d = Kd/dt
A1d = - 2.0*Kd/dt
A2d = Kd/dt
N := 5
tau := Kd / (Kp*N) // IIR filter time constant
alpha = dt / (2*tau)
d0 := 0
d1 := 0
fd0 := 0
fd1 := 0
loop:
 error[2] := error[1]
error[1] := error[0]
error[0] := setpoint − measured_value
// PI
output := output + A0 * error[0] + A1 * error[1]
// Filtered D
d1 = d0
d0 = A0d * error[0] + A1d * error[1] + A2d * error[2]
fd1 = fd0
fd0 = ((alpha) / (alpha + 1)) * (d0 + d1) - ((alpha - 1) /
(alpha + 1)) * fd1
output := output + fd0
wait(dt)
goto loop

See also

Cont rol t heory

Notes

a. The only exception is where the target value is the same as the value obtained when the controller
output is zero.

b. A common assumption often made for Proportional-Integral-Derivative (PID) control design, as

done by Ziegler and Nichols, is to take the integral time constant to be four times the derivative
time constant. Although this choice is reasonable, selecting the integral time constant to have
this value may have had something to do with the fact that, for the ideal case with a derivative
term with no filter, the PID transfer function consists of two real and equal zeros in the
numerator.[20]

References

1. Araki, M. (2009). "CONTROL SYSTEMS, ROBOTICS AND AUTOMATION – Volume VII - PID Control"
(http://www.eolss.net/ebooks/Sample%20Chapters/C18/E6-43-03-03.pdf) (PDF) . Japan: Kyoto
University.

2. Hills, Richard L (1996), Power From the Wind, Cambridge University Press
 3. Richard E. Bellman (December 8, 2015). Adaptive Control Processes: A Guided Tour (https://book
s.google.com/books?
id=iwbWCgAAQBAJ&q=%22Centrifugal+Governor%22+Huygens&pg=PA36) . Princeton University
Press. ISBN 9781400874668.

4. Bennett, Stuart (1996). "A brief history of automatic control" (https://web.archive.org/web/201608

09050823/http://ieeecss.org/CSM/library/1996/june1996/02-HistoryofAutoCtrl.pdf) (PDF) . IEEE
Control Systems Magazine. 16 (3): 17–25. doi:10.1109/37.506394 (https://doi.org/10.1109%2F3
7.506394) . Archived from the original (http://ieeecss.org/CSM/library/1996/june1996/02-Histor
yofAutoCtrl.pdf) (PDF) on 2016-08-09. Retrieved 2014-08-21.

5. Maxwell, J. C. (1868). "On Governors" (https://upload.wikimedia.org/wikipedia/commons/b/b1/On

_Governors.pdf) (PDF) . Proceedings of the Royal Society. 100.

6. Newpower, Anthony (2006). Iron Men and Tin Fish: The Race to Build a Better Torpedo during
World War II. Praeger Security International. ISBN 978-0-275-99032-9. p. citing Gray, Edwyn
(1991), The Devil's Device: Robert Whitehead and the History of the Torpedo, Annapolis, MD: U.S.
Naval Institute, p. 33.

7. Sleeman, C. W. (1880), Torpedoes and Torpedo Warfare (https://archive.org/stream/torpedoestorp

edo00sleerich#page/136/mode/2up/search/depth) , Portsmouth: Griffin & Co., pp. 137–138,
"which constitutes what is termed as the secret of the fish torpedo."

8. "A Brief Building Automation History" (https://web.archive.org/web/20110708104028/http://www.

building-automation-consultants.com/building-automation-history.html) . Archived from the
original (http://www.building-automation-consultants.com/building-automation-history.html) on
2011-07-08. Retrieved 2011-04-04.

9. Minorsky, Nicolas (1922). "Directional stability of automatically steered bodies". Journal of the
American Society for Naval Engineers. 34 (2): 280–309. doi:10.1111/j.1559-3584.1922.tb04958.x
(https://doi.org/10.1111%2Fj.1559-3584.1922.tb04958.x) .

10. Bennett 1993, p. 67 (https://books.google.com/books?id=VD_b81J3yFoC&pg=PA67)

11. Bennett, Stuart (June 1986). A history of control engineering, 1800-1930. IET. pp. 142–148 (http
s://books.google.com/books?id=1gfKkqB_fTcC&pg=PA142) . ISBN 978-0-86341-047-5.

12. Shinskey, F Greg (2004), The power of external-reset feedback (https://classes.engineering.wustl.e

du/2009/spring/che433/2009-LAB/Control%20Theory/external-reset.pdf) (PDF) , Control Global

13. Neuhaus, Rudolf. "Diode Laser Locking and Linewidth Narrowing" (http://www.toptica.com/fileadm
in/user_upload/Articles_Application_Notes/toptica_AP_1012_laser_locking_2012.pdf) (PDF) .
Retrieved June 8, 2015.

14. "Position control system" (http://www.ee.hacettepe.edu.tr/~solen/ELE356/exp1.pdf) (PDF) .

Hacettepe University Department of Electrical and Electronics Engineering.
15. Kebriaei, Reza; Frischkorn, Jan; Reese, Stefanie; Husmann, Tobias; Meier, Horst; Moll, Heiko;
Theisen, Werner (2013). "Numerical modelling of powder metallurgical coatings on ring-shaped
parts integrated with ring rolling". Material Processing Technology. 213 (1): 2015–2032.
doi:10.1016/j.jmatprotec.2013.05.023 (https://doi.org/10.1016%2Fj.jmatprotec.2013.05.023) .

16. Lipták, Béla G. (2003). Instrument Engineers' Handbook: Process control and optimization
(4th ed.). CRC Press. p. 108. ISBN 0-8493-1081-4.

17. "Introduction: PID Controller Design" (http://ctms.engin.umich.edu/CTMS/index.php?example=Intr

oduction&section=ControlPID) . University of Michigan.

18. Tim Wescott (October 2000). "PID without a PhD" (http://igor.chudov.com/manuals/Servo-Tuning/

PID-without-a-PhD.pdf) (PDF) . EE Times-India.

19. Bechhoefer, John (2005). "Feedback for Physicists: A Tutorial Essay On Control". Reviews of
Modern Physics. 77 (3): 783–835. Bibcode:2005RvMP...77..783B (https://ui.adsabs.harvard.edu/
abs/2005RvMP...77..783B) . CiteSeerX 10.1.1.124.7043 (https://citeseerx.ist.psu.edu/viewdoc/s
ummary?doi=10.1.1.124.7043) . doi:10.1103/revmodphys.77.783 (https://doi.org/10.1103%2Fre
vmodphys.77.783) .

20. Atherton, Drek P (December 2014). "Almost Six Decades in Control Engineering". IEEE Control
Systems Magazine. 34 (6): 103–110. doi:10.1109/MCS.2014.2359588 (https://doi.org/10.1109%2
FMCS.2014.2359588) . S2CID 20233207 (https://api.semanticscholar.org/CorpusID:2023320
7) .

21. Li, Y., et al. (2004) CAutoCSD - Evolutionary search and optimisation enabled computer
automated control system design, Int J Automation and Computing, vol. 1, No. 1, pp. 76-88.
ISSN 1751-8520 (https://www.worldcat.org/search?fq=x0:jrnl&q=n2:1751-8520) .

22. Kiam Heong Ang; Chong, G.; Yun Li (2005). "PID control system analysis, design, and technology"
(http://eprints.gla.ac.uk/3817/1/IEEE3.pdf) (PDF) . IEEE Transactions on Control Systems
Technology. 13 (4): 559–576. doi:10.1109/TCST.2005.847331 (https://doi.org/10.1109%2FTCST.2
005.847331) . S2CID 921620 (https://api.semanticscholar.org/CorpusID:921620) .

23. Jinghua Zhong (Spring 2006). "PID Controller Tuning: A Short Tutorial" (https://web.archive.org/w
eb/20150421081758/http://saba.kntu.ac.ir/eecd/pcl/download/PIDtutorial.pdf) (PDF) . Archived
from the original (http://saba.kntu.ac.ir/eecd/pcl/download/PIDtutorial.pdf) (PDF) on 2015-04-
21. Retrieved 2011-04-04.

24. Åström, K.J.; Hägglund, T. (July 1984). "Automatic Tuning of Simple Regulators" (https://lup.lub.lu.
se/record/8601786) . IFAC Proceedings Volumes. 17 (2): 1867–1872. doi:10.1016/S1474-
6670(17)61248-5 (https://doi.org/10.1016%2FS1474-6670%2817%2961248-5) .

25. Hornsey, Stephen (29 October 2012). "A Review of Relay Auto-tuning Methods for the Tuning of
PID-type Controllers" (http://www2.warwick.ac.uk/fac/cross_fac/iatl/reinvention/issues/volume5i
ssue2/hornsey) . Reinvention. 5 (2).

26. Bequette, B. Wayne (2003). Process Control: Modeling, Design, and Simulation. Upper Saddle
River, New Jersey: Prentice Hall. p. 129. ISBN 978-0-13-353640-9.
27. Heinänen, Eero (October 2018). A Method for automatic tuning of PID controller following Luus-
Jaakola optimization (https://dspace.cc.tut.fi/dpub/bitstream/handle/123456789/26690/Heinan
en.pdf) (PDF) (Master's Thesis ed.). Tampere, Finland: Tampere University of Technology.
Retrieved Feb 1, 2019.

28. Li, Yun; Ang, Kiam Heong; Chong, Gregory C.Y. (February 2006). "Patents, software, and hardware
for PID control: An overview and analysis of the current art" (http://eprints.gla.ac.uk/3816/1/IEEE
2pdf.pdf) (PDF) . IEEE Control Systems Magazine. 26 (1): 42–54.
doi:10.1109/MCS.2006.1580153 (https://doi.org/10.1109%2FMCS.2006.1580153) .
S2CID 18461921 (https://api.semanticscholar.org/CorpusID:18461921) .

29. Soltesz, Kristian (January 2012). On Automation of the PID Tuning Procedure (http://www.control.l
th.se/documents/2012/sol2012lic.pdf) (PDF) (Licentiate theis). Lund university. 847ca38e-
93e8-4188-b3d5-8ec6c23f2132 (http://lup.lub.lu.se/record/2293573) .

30. Li, Y. and Ang, K.H. and Chong, G.C.Y. (2006) PID control system analysis and design - Problems,
remedies, and future directions (http://eprints.gla.ac.uk/3815/1/IEEE_CS_PID_01580152.pdf) .
IEEE Control Systems Magazine, 26 (1). pp. 32-41. ISSN 0272-1708 (https://www.worldcat.org/se
arch?fq=x0:jrnl&q=n2:0272-1708)

31. Cooper, Douglas. "Integral (Reset) Windup, Jacketing Logic and the Velocity PI Form" (http://www.
controlguru.com/2008/021008.html) . Retrieved 2014-02-18.

32. Cooper, Douglas. "PI Control of the Heat Exchanger" (http://www.controlguru.com/wp/p71.html) .

Practical Process Control by Control Guru. Retrieved 2014-02-27.

33. Yang, T. (June 2005). "Architectures of Computational Verb Controllers: Towards a New Paradigm
of Intelligent Control". International Journal of Computational Cognition. 3 (2): 74–101.
CiteSeerX 10.1.1.152.9564 (https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.152.95
64) .

34. Liang, Yilong; Yang, Tao (2009). "Controlling fuel annealer using computational verb PID
controllers" (http://dl.acm.org/citation.cfm?id=1719210&preflayout=tabs) . Proceedings of the
3rd International Conference on Anti-Counterfeiting, Security, and Identification in
Communication. Asid'09: 417–420. ISBN 9781424438839.

35. Tenreiro Machado JA, et al. (2009). "Some Applications of Fractional Calculus in Engineering" (htt
ps://doi.org/10.1155%2F2010%2F639801) . Mathematical Problems in Engineering. 2010: 1–34.
doi:10.1155/2010/639801 (https://doi.org/10.1155%2F2010%2F639801) . hdl:10400.22/4306 (h
ttps://hdl.handle.net/10400.22%2F4306) .

36. [1] (https://www.controleng.com/articles/fundamentals-of-cascade-control/) Fundamentals of

cascade control | Sometimes two controllers can do a better job of keeping one process variable
where you want it. | By Vance VanDoren, PHD, PE | AUGUST 17, 2014

37. [2] (https://www.watlow.com/blog/posts/benefits-of-cascade-control) | The Benefits of Cascade

Control | September 22, 2020 | Watlow
38. King, Myke (2011). Process Control: A Practical Approach. Wiley. pp. 52–78. ISBN 978-0-470-
97587-9.

39. "Discrete PI and PID Controller Design and Analysis for Digital Implementation" (https://www.scrib
d.com/doc/19070283/Discrete-PI-and-PID-Controller-Design-and-Analysis-for-Digital-Implement
ation) . Scribd.com. Retrieved 2011-04-04.

40. "PID process control, a "Cruise Control" example" (http://www.codeproject.com/Articles/36459/PI

D-process-control-a-Cruise-Control-example) . CodeProject. 2009. Retrieved 4 November 2012.

Bequette, B. Wayne (2006). Process Control: Modeling, Design, and Simulation. Prentice Hall PTR.
ISBN 9789861544779.

Further reading

Liptak, Bela (1995). Instrument Engineers' Handbook: Process Control. Radnor, Pennsylvania: Chilton
Book Company. pp. 20–29. ISBN 978-0-8019-8242-2.

Tan, Kok Kiong; Wang Qing-Guo; Hang Chang Chieh (1999). Advances in PID Control. London, UK:
Springer-Verlag. ISBN 978-1-85233-138-2.

King, Myke (2010). Process Control: A Practical Approach (http://eu.wiley.com/WileyCDA/WileyTitle/

productCd-0470975873.html) . Chichester, UK: John Wiley & Sons Ltd. ISBN 978-0-470-97587-9.

Van Doren, Vance J. (July 1, 2003). "Loop Tuning Fundamentals" (http://old.controleng.com/articl

e/268148-Loop_ Tuning_ Fundamentals.php.html) . Control Engineering.

Sellers, David. "An Overview of Proportional plus Integral plus Derivative Control and Suggestions
for Its Successful Application and Implementation" (https://web.archive.org/web/2007030716174
1/http://www.peci.org/library/PECI_ ControlOverview1_ 1002.pdf) (PDF). Archived from the original
(http://www.peci.org/library/PECI_ ControlOverview1_ 1002.pdf) (PDF) on March 7, 2007. Retrieved
2007-05-05.

Graham, Ron; Mike McHugh (2005-10-03). "FAQ on PID controller tuning" (https://web.archive.org/w
eb/20050206113949/http://www.tcnj.edu/~rgraham/PID-tuning.html) . Mike McHugh. Archived
from the original (http://www.tcnj.edu/~rgraham/PID-tuning.html) on February 6, 2005. Retrieved
2009-01-05.

Aidan O'Dwyer (2009). Handbook of PI and PID Controller Tuning Rules (http://cyxtp.ucoz.ru/pdf/Aid
an_ O_ Dwyer_ Handbook_ of_ PI_ and_ PID_ Controller_ Tuning_ Rules.pdf) (PDF) (3rd ed.). Imperial
College Press. ISBN 978-1-84816-242-6.

External links

PID t uning using Mat hemat ica (ht t p://reference.wolfram.com/mat hemat ica/ref/PIDTune.h
t ml)
 PID t uning using Pyt hon (ht t ps://apmonit or.com/pdc/index.php/Main/Proport ionalInt egralD
erivat ive)

Principles of PID Cont rol and Tuning (ht t p://www.eurot herm.com/t emperat ure-cont rol/prin
ciples-of-pid-cont rol-and-t uning)

Int roduct ion t o t he key t erms associat ed wit h PID Temperat ure Cont rol (ht t p://www.eurot
herm.com/t emperat ure-cont rol/pid-cont rol-made-easy)

PID tutorials
PID Cont rol in MATLAB/Simulink and Pyt hon wit h TCLab (ht t p://apmonit or.com/pdc/index.
php/Main/ArduinoTemperat ureCont rol)

What 's All This P-I-D St uff, Anyhow? (ht t p://elect ronicdesign.com/analog/what s-all-p-i-d-s
t uff-anyhow) Art icle in Elect ronic Design

Shows how t o build a PID cont roller wit h basic elect ronic component s (ht t p://cit eseerx.ist .
psu.edu/viewdoc/download?doi=10.1.1.154.240&rep=rep1&t ype=pdf) (pg. 22)

PID Wit hout a PhD (ht t ps://www.wescot t design.com/art icles/pid/pidWit hout APhd.pdf)

PID Cont rol wit h MATLAB and Simulink (ht t p://www.mat hworks.com/discovery/pid-cont rol.
ht ml)

PID wit h single Operat ional Amplifier (ht t p://www.post reh.com/vmichal/papers/PID-Radio.p

df)

Proven Met hods and Best Pract ices for PID Cont rol (ht t p://www.cont rolguru.com/pages/t
able.ht ml)

Principles of PID Cont rol and Tuning (ht t p://www.eurot herm.com/t emperat ure-cont rol/prin
ciples-of-pid-cont rol-and-t uning)

PID Tuning Guide: A Best -Pract ices Approach t o Underst anding and Tuning PID Cont rollers
(ht t p://cont rolst at ion.com/port folio-it ems/pid-t uning-guide-a-best -pract ices-approach-t o
-underst anding-and-t uning-pid-cont rollers/?port folioID=7052)

Michael Barr (2002-07-30), Introduction to Closed-Loop Control (ht t ps://web.archive.org/we

b/20100209125831/ht t p://embedded.com/st ory/OEG20020726S0044) , Embedded
Syst ems Programming, archived from t he original (ht t ps://www.embedded.com/st ory/OEG
20020726S0044) on 2010-02-09

Jinghua Zhong, Mechanical Engineering, Purdue Universit y (Spring 2006). "PID Cont roller
Tuning: A Short Tut orial" (ht t ps://web.archive.org/web/20150421081758/ht t p://saba.knt u.a
c.ir/eecd/pcl/download/PIDt ut orial.pdf) (PDF). Archived from t he original (ht t p://saba.knt
u.ac.ir/eecd/pcl/download/PIDt ut orial.pdf) (PDF) on 2015-04-21. Ret rieved 2013-12-04.
 Int roduct ion t o P,PI,PD & PID Cont roller wit h MATLAB (ht t p://vivekbose.com/int roduct ion-
t o-pid-cont roller-wit h-det ailed-ppipd-pd-cont rol)

Online calculators

Wikimedia Commons has media related to PID controllers .

Online PID t uning applicat ion (ht t ps://pidt uner.com)

Retrieved from
"https://en.wikipedia.org/w/index.php?
title=PID_controller&oldid=1121264197"

Last edited 20 days ago by 88.12.150.50