what is the purpose of having a stable instrument in process control? By stable instrument I mean instrument that gives consistent measurement to the control system.

 

The first question to ask is why the signal should be unstable?

Is it because the process upstream conditions are variable? e.g. one or more component streams show signs of quality variation.
If so, then that is why you need control which is to respond to such changes and thus which needs the sensor to be responsive to such changes.

Is it because the instrument is too sensitive? Is it correctly specified and installed? If the quality is stable then the instrument reading need also be stable. If not, why not?
Is it because it is not operating at its optimum conditions? Is installed properly?
If the process parameter is stable then we need the instrument to generate a stable but responsive signal.
If the process parameter is varying then the control system needs to respond to correct for the excursion and thus the instrument needs to show this variation by responding quickly and accurately. But we may want to be able to ignore normally expected noise.

Each item in the control loop will have a response time and various factors need to be considered in how the controller is tuned to respond to any deviation. e.g. the sensor, the controller, the control valve and so on.

In some cases fluctuating parameter can lead to a control response that, due to the response time, takes effect just as the parameter changes back to its original value.
In such a case the control response introduces an excursion and the system may begin to hunt.

In an NaOH blending system I just witnessed, the system was set up and run in the manual mode.
The sensor is a tube densitometer which determines the density at a reference temperature and then converts this to concentration.
In manual mode the operation was entirely stable and the density meter readings and the calculation of concentration were both highly stable with very minor variations in the observed density.

When the system was first put under automatic control the density and concentration readings were seen to be more unstable.
This was because the PID controller was responding to the minor fluctuations and it was then tuned such that the small fluctuations in the reading normally evident were effectively ignored.

So why have a PID control if the system is stable anyway?
Because as the raw caustic quality may vary needing more or less water to maintain the target value.

In this case we have a sensitive instrument with a fast response time. That means that the control system can respond quickly to excursions and thus maintain precise control over the quality while not being affected by minor fluctuations.

In this system also the control valve, when the system was operating at the target quality, was approximately in mid point and evidently the balance of the raw caustic and the water flow streams was stable as shown when the valve was not modulated but set to a single position under manal control.

If the valve had not been at its optimum control position then there might be fluctuations in the flows of the component streams. In that case the sensor would necessarily show a fluctuation in the measurement.

Where the measurement is unstable or shows a degree of fluctuation that can cause difficulties in establishing stable control, one approach is to "smooth" the signal output.
Some systems use "capacitive" style smoothing or simple averaging.
This means that in any calculation style the last N readings are collected, the highest and lowest rejected and the remaining values averaged. AT each update the oldest reading is discarded and the newest added to the stack.
This may result in a stable signal insensitive to fluctuations and hence unlikely to cause the valve to hunt, but it may also mean that the signal is not so responsive to a genuine change in the quality.

I personally always like to be able to see a raw sensor measurement because it reveals something about what is going on. In cases where the signal is expected to be noisy but where step changes in quality may be experienced, then the "smoothing algorithm needs to be more sophisticated using, for example, step alarms that over-ride the averaging.