Manual PID Tuning Step By Step

Most modern PID controllers will tune themselves. Auto-tune observes how your process responds to a step change and calculates the parameters for you, and on a well-behaved heating system it usually gets close enough to leave alone. We recommend running auto-tune first on any new installation.

There are situations where it is not enough. A process with a long lag, a load that changes significantly between runs, or a system where auto-tune has been run and the result still overshoots or hunts, may need the parameters setting by hand. This page covers two manual methods and, more usefully, how to translate the numbers they produce into the parameters a Novus controller actually asks for.

Before You Start

Both methods work by deliberately driving the process into oscillation. Note the below precautions:

Do not use these methods on a process where temperature overshoot could damage product, equipment or people. Chemical processes, anything with a flammable load, and anything with a person nearby are the obvious cases.
Set a sensible maximum power limit before you begin, and make sure any over-temperature protection is working and independent of the controller you are tuning.
Tune with the real load fitted, from a cold start, with no other disturbances running. Parameters found on an empty oven will not suit a full one.
Write down the existing parameters first. If manual tuning goes badly you want to be able to put the controller back as it was.

It is also worth checking the basics before assuming the parameters are at fault. A sensor in the wrong place, or an input type that does not match the sensor fitted, produces behaviour that looks exactly like a tuning problem and will not be fixed by tuning.

Method 1: Ziegler-Nichols Closed Loop

The Ziegler-Nichols closed loop method, published by John Ziegler and Nathaniel Nichols in 1942, is the standard reference procedure and still the one most engineers reach for. It finds the point at which the process becomes marginally stable, then backs off from it by a set of fixed ratios.

Set the controller to a setpoint typical of normal operation.
Turn integral and derivative action off, leaving proportional only.
Start with a low proportional gain and increase it in steps. After each step, nudge the setpoint by around 5% and watch the response.
Keep going until the process oscillates steadily - neither growing nor dying away. The gain at that point is the ultimate gain, which we will call Ku. The time for one complete cycle of the oscillation is the ultimate period, Tu, in seconds.
Note both values, then set the parameters from the table below.

Ziegler-Nichols parameters, from Ku and Tu:

Control Type	Gain (Kc)	Integral Time (Ti)	Derivative Time (Td)
P only	0.50 x Ku	-	-
PI	0.45 x Ku	Tu / 1.2	-
PID	0.60 x Ku	Tu / 2	Tu / 8

Ziegler-Nichols is tuned for disturbance rejection rather than a smooth approach to setpoint, so it generally leaves the process slightly underdamped - expect some overshoot on the first run. If that matters, halving the gain is the usual first adjustment.

Method 2: Reduced Gain

The second method reaches a similar place with less oscillation, which makes it the safer choice on a process you would rather not push hard.

As before, set a typical setpoint, turn integral and derivative off, and set a safe maximum power.
Increase the proportional gain until the process is just oscillating, then reduce it by about 30%. The oscillation should stop.
Nudge the setpoint by around 5% to provoke a response. Add derivative action and increase it until the process settles without overshooting - as a rule of thumb, around one third of the oscillation period you saw in step 2 is a reasonable starting point.
Increase integral action slowly until oscillation just begins, then reduce it by a factor of two or three. Use the least integral action that removes the steady-state offset; more than that tends to cause problems later.
Run the process through a normal cycle and confirm it behaves under real conditions.

Converting The Numbers For A Novus Controller

This is the step that catches people out. The tuning literature works in gain and integral time. Novus controllers, in common with most industrial temperature controllers, ask for proportional band, integral rate and derivative time instead. The numbers from the table above need converting before you enter them.

Proportional Band (Pb) is the inverse of gain, expressed as a percentage of the input span: Pb = 100 / Kc. A gain of 4 is a proportional band of 25%. Note that increasing Pb makes the controller less aggressive, which is the opposite of increasing gain.
Integral Rate (Ir) on a Novus controller is expressed in repeats per minute, where the tuning tables give integral time in seconds: Ir = 60 / Ti. An integral time of 120 seconds is an integral rate of 0.5.
Derivative Time (Dt) is in seconds and transfers directly from the table. On Novus controllers we normally see quite small values here.

Parameter names and units do vary between models, so it is worth confirming against the manual for your specific controller before entering anything. Manuals are available on each product page.

If Tuning Does Not Fix It

Manual tuning solves a tuning problem. It will not solve a hardware or installation problem, and a good deal of what gets reported to us as a tuning fault turns out to be something else.

If the process still misbehaves after a careful manual tune, the usual causes are sensor position, an input type that does not match the sensor fitted, or a mechanical relay being asked to switch faster than it comfortably can. Those are covered in more detail in the guides below.

Frequently Asked Questions

Should I tune manually or use auto-tune?
Auto-tune first, in almost every case. Manual tuning is worth the time when auto-tune has been run under representative conditions and the result still overshoots or hunts, or where the process cannot safely be put through an auto-tune cycle.

How long should the whole process take?
On a small oven, under an hour. On something with a lot of thermal mass it can take considerably longer, simply because each gain step needs enough time for the process to show what it is doing before you change anything.

My controller only shows Pb, Ir and Dt - where do I enter gain?
You do not. Convert the gain to a proportional band using Pb = 100 / Kc, as above. This is the most common reason manual tuning gives an unexpected result on a first attempt.

Do I need to re-tune if I change the heater or the load?
Yes. Parameters describe how a particular process responds, so a different heater, a different enclosure, added thermal mass or a significant change in ambient conditions all justify a fresh tune.

Can I copy parameters from an identical machine?
As a starting point, often yes, and it will usually save time. Confirm the behaviour on the second machine rather than assuming it, since small differences in sensor position and airflow have a larger effect than people expect.

Shop Related Equipment

Shop the full range of PID Temperature Controllers, or see the Solid State Relays often used to give PID finer control of a heating element.