# PID controller explained: how it works, settings and process dynamics > How does a PID controller work? Learn what P, I and D do, how process dynamics determine the right settings and how to recognise control behaviour in trend data. Published: 2026-08-17 Updated: 2026-09-19 A PID controller is the link between what a process should do and what the process actually does. The desired value is the setpoint (SP), the measured process value is the PV, and the controller output (MV or output) is used to correct the process. The difference between SP and PV is the control error the controller responds to. What that correction looks like is determined by the P, I and possibly D action. But good PID control does not start with adjusting Kp, Ti or Td. The settings have to match the dynamics of the process behind them. Dead time, process gain, time constants and disturbances ultimately determine how fast and how aggressively a control loop can respond. That is why a set of PID settings says little on its own. To judge whether a PID loop is set well, you have to look at the controller and at the behaviour of the process. ## What is a PID controller? A PID controller is a feedback controller that keeps a process value at its setpoint by combining three actions: proportional (P), integral (I) and derivative (D). It works continuously in a [closed control loop](/en/knowledge/process-control#how-is-a-control-loop-structured). In such a PID control loop, the same four steps are repeated again and again: **Measure.** A transmitter measures the process value (PV), for example pressure, flow, level or temperature. **Compare.** The controller compares the PV with the setpoint (SP). The difference between the two is the control error. **Correct.** Based on that control error, the controller calculates how much the controller output (MV or output) needs to change. **Measure again.** The adjustment affects the process, the PV responds and the controller assesses the error again. This feedback is the basis of every PID loop. The PID controller responds to the difference between the desired and the measured process value, but how the process responds to a correction is determined by the process dynamics. That is why the same PID settings can produce completely different control behaviour in two different processes. ## What P, I and D stand for in a PID control loop #### P: the proportional action The proportional action responds directly to the current control error between setpoint and process value. The larger this error, the larger the controller’s correction. The strength of this response is determined by the proportional gain Kp. A higher Kp makes the controller respond more strongly to the same error. This can make the PID loop faster and more accurate, but too high a gain can lead to overshoot and oscillations. With a low Kp the controller responds less strongly, so the loop can become sluggish and errors are corrected more slowly. With P action alone, many processes are left with a persistent difference between PV and setpoint. A constant controller output requires a certain control error. This is called the steady-state error or offset. The I action makes sure this persistent error is eventually removed. #### I: the integral action The integral action does not only look at the current control error, but builds up correction for as long as there is a difference between setpoint (SP) and process value (PV). This allows the I action to eventually remove the steady-state error that can remain with P action alone. The speed at which this correction builds up is determined by the integral time Ti. Depending on the control system, it is also called Tn, reset time or integral action time. The integral time is usually expressed in seconds or minutes. It is important to note that Ti can feel counter-intuitive: a smaller Ti means a stronger and faster I action. A larger Ti makes the integral action calmer and slower. If Ti is too small, the integral action can build up correction faster than the process can handle. The controller keeps correcting while the effect of earlier corrections is not yet fully visible, which can cause overshoot and oscillations. If Ti is too large, the integral action becomes sluggish and it can take a long time before a persistent control error has been fully removed. #### D: the derivative action The derivative action responds to how fast the control error is changing. Where P looks at the current error and I at the error accumulated over time, D looks at the rate of change. The strength of this action is determined by the derivative time Td, also called Tv. When the PV moves quickly towards the setpoint, the D action can brake the correction earlier. In that way D can add extra damping and limit overshoot. This can be especially valuable in processes where the process value changes predictably and gradually and contains enough information about the direction in which the process is developing. D action does have an important drawback, however. Fast changes and noise in the measurement signal can be amplified by the derivative action. This can make the controller output restless and make a valve or other actuator move unnecessarily. That is why D is not used in many industrial control loops and a PI controller is chosen instead. Whether D adds value is therefore not only determined by how fast or slow a process is. The dynamic behaviour of the process, the quality of the measurement and the effect on the controller output are decisive too. D action should therefore only be applied when it demonstrably adds damping or improves control behaviour. ## How process dynamics determine the PID settings The right settings of a PID loop are not only determined by the controller, but above all by the process behind it. The same settings can therefore cause completely different control behaviour in two different processes. To determine how a process responds, three properties are particularly important: the process gain Kproc, the time constant τ and the dead time θ. Together they describe how strongly, how fast and with how much delay a process responds to a change in the controller output. #### Process gain Kproc The process gain Kproc indicates how strongly the process value responds to a change in the controller output. In simple form: *Kproc = ΔPV / ΔMV* Here ΔMV is the change in the controller output and ΔPV the change in the process value after the process has reached a new steady state. Suppose the controller output changes from 40 to 50 percent and the PV eventually rises from 100 to 120 as a result. The change in MV is then 10 percentage points and the change in PV 20 units: *Kproc = 20 / 10 = 2* Under these conditions, one percentage point of change in the controller output therefore eventually causes about two units of change in the PV. Kproc thus says something about the sensitivity of the process. A high process gain means that a relatively small change in the controller output can have a large effect on the PV. With a low process gain, more change in the controller output is needed for the same effect. The direction matters too. With a positive process gain, MV and PV move in the same direction. With a negative process gain, an increase in MV causes the PV to fall. Kproc does not yet tell you how fast that change takes place, however. For that we look at the time constant. #### Time constant τ The time constant τ describes how fast the process responds once a change has started. Many industrial processes do not reach a new final value immediately after a change in the controller output. The PV moves gradually towards a new state. For processes that behave roughly like a first-order system, the time constant can be estimated with the 63.2 percent method. After one time constant, the PV has covered about 63.2 percent of the total change towards the new final value. If the PV eventually rises from 100 to 120, for example, the total change is 20 units. The point at 63.2 percent of this change is at: *100 + (0.632 × 20) = 112.64* The time the process needs, from the start of the actual response, to reach this point gives an approximation of τ. A small τ indicates a relatively fast process response. A large τ means the process needs more time to respond to a change. But sometimes, after a change in the controller output, practically nothing happens at first. We describe that delay with the dead time. #### Dead time θ The dead time θ is the time between a change in the controller output and the moment its effect starts to become visible in the PV. Dead time can be caused, for example, by transport through a pipe, mixing, heat transfer, measurement delay or physical distance between the point where action is taken and the point where measurement takes place. For a PID controller, dead time matters. During this period the controller has no visible feedback yet from its earlier correction. The controller output has already changed, while the PV has not yet responded, or hardly. If the controller keeps correcting further during that period, it may later turn out that it corrected more than necessary. A process with relatively much dead time therefore usually calls for more cautious tuning than a process whose effect becomes visible quickly. #### Kproc, τ and θ together determine process behaviour The three properties each describe a different part of the same process response: *Kproc tells you how strongly the process responds.* *τ tells you how fast the response develops.* *θ tells you how much delay there is before that response becomes visible.* The ratio between dead time and time constant is particularly important. When θ is large relative to τ, the controller works for a relatively long time with information in which the effect of its latest corrections is not yet visible. That reduces the room for aggressive tuning. This does not mean that one value on its own determines whether a process is easy or hard to control. It is precisely the combination of process gain, speed and delay that determines which control settings suit the process. #### From trend data to process dynamics Kproc, τ and θ are not just theoretical concepts. Under suitable conditions these properties can be derived from the behaviour visible in process data. A change in the controller output and the subsequent response of the PV contain information about how strongly, how fast and with how much delay the process responds. That allows process behaviour to be described quantitatively instead of only being judged visually. That forms the basis for the next step: translating process behaviour into a process model and ultimately into suitable control settings. In our [results](/en/results) you can see how PID Dynamics applies these principles in practice, with the figures from before and after the optimisation. ## How to recognise a poorly performing PID loop in trend data Many problems in a control loop leave recognisable patterns in historical trend data. By looking at setpoint (SP), process value (PV) and controller output (MV or output) over the same period, a first picture of the control behaviour often emerges. A trend alone does not directly prove what the cause is, however. The same pattern can arise from wrong control settings, process disturbances, actuator problems or interactions with other control loops. Trend analysis is therefore the start of the diagnosis, not the end. #### The loop responds too aggressively A common pattern is a PV that repeatedly oscillates around the setpoint, while the controller output keeps correcting. When this movement has a recognisable rhythm and persists for a longer time, it can point to a control loop that is tuned too aggressively. Too strong a P action, too strong an I action or a combination of the two can make the controller correct more strongly each time than the process needs. Oscillation alone is not enough to draw this conclusion, however. A periodic process disturbance, interaction with another control loop or mechanical behaviour of an actuator can cause a similar pattern. That is why PV and controller output must always be assessed in relation to the process. #### The loop responds too slowly The opposite happens too. After a change or disturbance the PV does move back towards the setpoint, but it takes relatively long before the error is corrected. In trend data this can show up as a long-lasting control error while the controller output responds only to a limited extent or slowly. This can fit a conservatively tuned controller, for example due to a low proportional gain or weak integral action. But a slow response does not automatically mean the PID settings are wrong. The natural process dynamics, limits or insufficient capacity of the final control element can also make the PV recover slowly. The speed of the process must therefore always be taken into account before the control settings are changed. #### The integral action runs into a limit A controller output has physical limits. A valve cannot open further than 100 percent, for example, and a pump cannot deliver unlimited extra capacity. When the controller output reaches its minimum or maximum value, the controller can no longer correct the process further in the same direction. If the integral action keeps building up correction in the meantime, integrator windup can occur. As soon as the process conditions change or the controller output has room to move again, this accumulated integral contribution first has to be wound down. This can lead to a long recovery or a large overshoot. In trend data this can be recognisable, for example, as a controller output that sits at a limit for a longer time, combined with a PV that then moves well past the setpoint. Modern controllers often have anti-windup functions to limit this behaviour. Still, prolonged saturation of the controller output remains important information when analysing a control loop. #### The D action responds to measurement noise The derivative action responds to fast changes in the measurement signal. As a result, small, fast fluctuations in the PV can cause relatively strong movements of the controller output. On a zoomed-out trend the PV can look reasonably calm, while zooming in reveals small, fast changes. When these movements come back amplified in the controller output, the D action may be playing a part. This can lead to a restless controller output and unnecessary movement of valves or other actuators. In such a situation it should be investigated whether the D action actually adds value and whether the quality and filtering of the measurement signal are suitable for derivative action. #### And sometimes it is not the PID settings at all Not every oscillation, slow response or persistent error is caused by Kp, Ti or Td. A control valve can suffer from stiction, for example. Due to static friction, the signal changes but the valve hardly moves at first. As soon as the friction is overcome, the valve can suddenly jump. This can cause a recurring pattern in both the process value and the controller output. Saturation, measurement problems, process disturbances, changing operating conditions and interactions between different control loops can also cause behaviour that looks like wrong PID settings. That is why looking only at the shape of the PV is not enough. The relationship between SP, PV, controller output and the process around it ultimately determines what is really happening. That distinction is essential. Retuning a controller while the real cause lies elsewhere does not solve the problem and can even make the control behaviour worse. ## Why a PID loop can perform differently over time A PID controller can keep running for years with the same Kp, Ti and possibly Td. Yet that does not mean the loop still performs the same after years as it did with the original tuning. The reason is simple: the control settings stay the same, while the dynamics of the process can change. Heat exchangers can foul, valves can wear or get a different characteristic, and pumps can respond differently after maintenance or replacement. Changes in production capacity, raw materials, product mix, temperatures, pressures or other operating conditions can also affect the behaviour of a process. With that, properties such as the process gain Kproc, time constant τ and dead time θ can change too. The Kp and Ti that suited the original process dynamics need no longer be optimal for the current situation. This does not have to lead directly to a fault or alarm. A control loop can keep functioning technically while the PV varies more, disturbances are removed more slowly or the controller output moves more than necessary. That is exactly why deteriorating control performance can go unnoticed for a long time. Good tuning should therefore not be seen only as something done once during commissioning. When the process, the plant or the operating conditions change, it makes sense to reassess whether the control settings still match the current process dynamics. ## How to tune a PID controller properly There are various ways to tune a PID controller. The main difference lies in what the chosen settings are based on. #### Tuning by experience In practice, PID settings are regularly adjusted by hand based on experience and the visible behaviour of the loop. Kp or Ti is changed, after which it is judged whether the trend improves. An experienced operator or engineer can achieve good results this way. The drawback is that the outcome depends heavily on experience, available time and the conditions during testing. In addition, it is not always clear afterwards why certain settings were chosen. #### Classic tuning methods A well-known example is the Ziegler Nichols method. In the classic closed-loop method, the proportional gain is increased until a sustained oscillation occurs. Control settings can then be determined from the critical gain and oscillation period. This method played an important role in the development of PID tuning, but deliberately pushing a control loop towards the stability limit is not desirable for every industrial plant. There are therefore various other tuning methods that arrive at suitable control settings in a different way. #### Tuning based on process dynamics Another approach is to first determine the dynamic behaviour of the process. The process gain Kproc, time constant τ and dead time θ discussed earlier can, under suitable conditions, be estimated from the response of the process. With these process parameters, a model of the process can be built. Based on that, suitable values for Kp, Ti and, where needed, Td can then be calculated. The big difference is that the control settings are then not chosen solely based on how a trend looks. There is a technical relationship between the measured process behaviour and the chosen tuning. Historical trend data can be valuable here because existing process movements can be used to examine the dynamics, without automatically requiring new tests on the plant. The available data must, however, contain enough information to draw reliable conclusions. PID Dynamics works from this approach. We first analyse the control behaviour and the underlying process dynamics and, based on that, substantiate which adjustments make sense. How that works, from the first analysis to implementation and verification, can be read in [the full programme](/en/services). ## Frequently asked questions #### What is a PID controller? A PID controller is a control algorithm that calculates the controller output based on the difference between the desired value, the setpoint (SP), and the measured process value (PV). The control can use a proportional, integral and derivative action. In the process industry, PID controllers are used among other things to control pressure, flow, level, temperature and dosing. In many applications only P and I are used, so the controller works as a PI controller. #### What does PID stand for? PID stands for Proportional, Integral and Derivative. The P action responds to the current control error. The I action builds up correction for as long as an error remains. The D action responds to the rate at which the control error changes. Together these actions determine how the controller responds to changes and disturbances in the process. #### What is the difference between a PID controller and a PID control loop? The PID controller is the algorithm that performs the control calculation, for example as a function block in a DCS or PLC. The PID control loop, or [control loop](/en/knowledge/process-control) in general, comprises more than just that algorithm. The measurement, the actuator, the process and the feedback to the controller are part of the loop too. That distinction is important. Poor control behaviour does not have to be caused by wrong Kp, Ti or Td settings. Process disturbances, measurement problems, actuator behaviour, limits and interactions with other control loops can also play a role. #### What is the difference between a PI and a PID controller? A PI controller only uses the proportional and integral action. A PID controller also uses the derivative action. In many industrial control loops, PI control is sufficient. D action can add extra damping when the process dynamics are suitable for it and the measurement signal is of sufficient quality. D action does respond strongly to fast changes in the measurement signal, however, and can therefore also amplify measurement noise. Adding D is therefore not a standard improvement. The choice between PI and PID should match the behaviour of the process and the quality of the measurement. #### Can I tune a PID controller myself? You can, but a change in Kp, Ti or Td must suit the dynamics of the process. Tuning by hand based on experience can work well for certain control loops. As dead time, process interactions, disturbances or changing operating conditions play a bigger role, however, it becomes harder to determine which settings are suitable based solely on the visible trend. Substantiated tuning therefore starts with understanding the process. Parameters such as process gain Kproc, time constant τ and dead time θ can be used to describe the process dynamics and determine suitable control settings. #### How do I know whether poor control behaviour is caused by the PID settings? That cannot always be established from one trend or one symptom. An oscillating PV can arise from overly aggressive control settings, for example, but also from a periodic process disturbance, a valve with stiction or interaction with another control loop. A slow response can be the result of conservative tuning, but also of the natural process dynamics or a limit in the plant. That is why SP, PV and controller output are assessed together, preferably in combination with relevant process information. The pattern in the trend gives clues, but the underlying cause must be established first before the control settings are changed. #### How often should a PID loop be reassessed? There is no fixed interval for this. A well-tuned control loop does not automatically need to be retuned periodically. A new assessment becomes relevant mainly when the process dynamics or control performance change. Think of changed production conditions, fouling of equipment, maintenance or replacement of valves and pumps, changes in raw materials or a control loop that demonstrably starts to respond differently. The question is therefore not only how old the PID settings are, but above all whether they still suit the process as it functions today. ## Curious how much calmer your own loops can run? We analyse existing trend data and calculate what the optimal settings are. No extra hardware, no downtime. The PID Performance Scan is free: you first see where things can be calmer, and then receive a fixed price for the optimisation. [Schedule an introduction](/en/contact) and we will look at your situation together. You will hear from us within one to two working days.