# How do control loops perform in practice? What the research says

> Research into hundreds of thousands of PID control loops keeps showing the same picture: a large share is in manual, oscillates or runs into a limit. The figures, the sources and the caveats.

Published: 2026-10-07

A control loop that does not perform well rarely causes a fault. No alarm goes off and the process keeps running. That is why almost no plant knows exactly how its control loops are doing. Those who have measured it keep arriving at the same conclusion: a large share of control loops do not do what they should.

In this article we list the most important studies: what was measured, what came out of it and what to be careful with.

## What is a well-performing control loop?

A control loop performs well if it keeps the process close to the setpoint, damps disturbances quickly and does not move the valve more than necessary. It is in automatic and has control room left: the output does not run into 0 or 100%.

The studies below each use their own classification, but the signs they look for are always the same:

- **In manual.** The operator has switched the controller off and steers by hand.
- **At a limit.** The output sits at its minimum or maximum so often that the controller has nothing left to control.
- **Oscillating.** The process value oscillates around the setpoint, because of the tuning, a valve or another control loop.
- **More variation instead of less.** The controller makes the process more restless than it would be without a controller.

## Honeywell (2002): two thirds with clear room for improvement

Honeywell assessed 26,000 PID controllers in the continuous process industry, collected over two years. Each controller was placed in one of five classes, based on its variability and oscillation relative to what is achievable for that type of loop.

- 16% excellent
- 16% acceptable
- 22% fair
- 10% poor
- 36% open loop: the controller was in manual, or the output was at a limit, for more than 30% of the measurement period

The authors conclude that only a third of the controllers performed acceptably and two thirds had clear room for improvement. They also name a cause: a control engineer is typically responsible for two to four hundred control loops, alongside a lot of other work.

## Paulonis and Cox (2003): 41% fair or poor

Eastman Chemical built a system to assess the performance of its PID controllers automatically: flow, pressure, level and temperature loops in 40 plants at nine sites worldwide. Of those control loops, 41% fell into the fair or poor classes, mainly due to hardware problems: the valve, the positioner or the transmitter.

## Torres and colleagues (2006): 41% oscillate

Between July 2004 and October 2005, 700 control loops were audited at twelve companies in Brazil, in petrochemicals, pulp and paper, cement, chemicals, steel and mining. On average:

- 41% of the control loops oscillated, due to the tuning, coupling with other loops, disturbances or the actuator
- 24% had an output that sat at a limit most of the time
- 16% were in manual
- 16% had serious tuning problems
- 15% of the valves suffered from stiction or hysteresis
- 14% of the valves were badly worn

## ABB (2016): only a quarter improves the process

The largest dataset comes from ABB. Its monitoring system assesses about 600,000 PID control loops daily, at customers in a wide range of industries. Across those control loops the distribution was:

- 30% were in manual
- 30% increased the variability in the process
- 15% had an output outside its control range
- 25% improved the process

According to the authors, up to three quarters of the automation in a typical plant therefore does not work as designed, mainly because the loops have not been tuned thoroughly.

## What the studies show together

As early as 1993 the first audits came to the same conclusion. Bialkowski found in Canadian paper mills that only 20% of control loops worked well; in 30% the problem was the tuning and in another 30% the valves. Ender reported that 30% of controllers were in manual and 20% were still running on factory settings.

The studies differ in time, country and method, but the picture is remarkably stable. A 2024 doctoral thesis sums it up like this: across different industries and on different continents, about 40% of control loops do not achieve the performance expected of them.

The causes that keep coming back:

- **Tuned once.** Controllers are set during commissioning and rarely looked at again afterwards, while the process, the load and the valves change.
- **No time.** One control engineer has hundreds of control loops in their care. A loop that does not cause a fault does not make the list.
- **Valves and measurements.** Part of the unrest is not in the tuning, but in a valve that sticks or a measurement that is wrong.
- **Operators solve it themselves.** A loop that misbehaves goes to manual. That works, but afterwards it often stays there.

## What to be careful with

These figures say a lot about control loops in general, but nothing about your plant in particular. A few caveats:

- **Definitions differ.** "Open loop" means in manual or at a limit for Honeywell. The fair and poor classes are defined differently in each study. Do not add the percentages together and do not compare them one to one.
- **The data comes partly from suppliers.** Honeywell and ABB themselves sell software for monitoring control loops. Their figures have been published and peer-reviewed, but they also have an interest in the problem being visible.
- **Not everything in manual is wrong.** Some controllers are deliberately in manual, for example a bypass that is only used during start-up.
- **Older measurement periods.** The studies run from 1993 to 2016. That the picture hardly changed over that period does say something: better control systems do not solve it by themselves.

## How do you know how your control loops are doing?

It starts with looking at the trend data you already have. For each control loop, put the PV, the setpoint and the output side by side and look for the signs above: how often is it in manual, does the output run into a limit, does the PV oscillate?

For a first impression you can take the [quickscan](/en/tools/pid-quickscan): twelve questions about the signs of a restless control loop. If you want to see it substantiated per control loop, the free [PID Performance Scan](/en/services/performance-scan) uses your own trend data to show which loops are restless and why.

## Frequently asked questions

#### How many control loops are in manual on average?

That differs per study and per plant. ABB found that 30% of about 600,000 control loops were in manual, Torres and colleagues arrived at 16% for 700 control loops. Honeywell counted 36% as "open loop", but that also includes controllers whose output sits at a limit.

#### Why do operators put a controller in manual?

Usually because the loop does not do what they expect in automatic: it oscillates, responds too slowly or intervenes too hard during a load change. In manual they do have the process under control. The problem is that the loop often stays in manual afterwards, even when the cause could have been fixed long ago.

#### Is this still the case with modern DCS systems?

The studies show little improvement between 1993 and 2016, while control systems became much more powerful over that period. A modern DCS makes it easier to collect data, but the settings of a controller do not adapt by themselves when the process changes.

#### Is poor control behaviour always down to the tuning?

No. In the studies, valves with stiction or wear, incorrectly sized valves and measurement problems keep coming back as causes. That is why it is important to first establish where the unrest comes from, before the settings are changed. How to recognise that in the trend can be read in [PID controller explained](/en/knowledge/pid-controller-explained).

## Sources

- Desborough, L. and Miller, R. (2002). [*Increasing customer value of industrial control performance monitoring: Honeywell’s experience.*](https://skoge.folk.ntnu.no/prost/proceedings/cpc6-jan2002/desborough.pdf) AIChE Symposium Series 98(326).
- Starr, K.D., Petersen, H. and Bauer, M. (2016). [*Control loop performance monitoring: ABB’s experience over two decades.*](https://doi.org/10.1016/j.ifacol.2016.07.396) IFAC-PapersOnLine 49(7), 526-532.
- Paulonis, M.A. and Cox, J.W. (2003). *A practical approach for large-scale controller performance assessment, diagnosis, and improvement.* Journal of Process Control 13, 155-168. Figures as summarised by Jelali (2010).
- Torres, B.S., de Carvalho, F.B., de Oliveira Fonseca, M. and Filho, C.S. (2006). *Performance assessment of control loops: case studies.* IFAC ADCHEM, Gramado. Figures as summarised by Jelali (2010) and Yağcı and Arkun (2020).
- Jelali, M. (2010). [*Control System Performance Monitoring.*](https://duepublico2.uni-due.de/servlets/MCRFileNodeServlet/duepublico_derivate_00024588/Habil_Jelali.pdf) Habilitationsschrift, Universität Duisburg-Essen. Also contains the summary of Bialkowski (1993) and Ender (1993).
- Yağcı, M. and Arkun, Y. (2020). [*An integrated application of control performance assessment and root cause analysis in refinery control loops.*](https://doi.org/10.1016/j.ifacol.2020.12.651) IFAC-PapersOnLine 53(2), 11650-11655.
- Yağcı, M. (2024). *Control Performance Analysis with Industrial Scale Applications.* Doctoral thesis, Åbo Akademi.