# Means to improve precision and accuracy

I have sensor measuring a distance to a rotating disk. Later on I want to use these values to produce some results like the range of the values.

What options do I have to improve accuracy and precision of the results?

Things I've come up so far:

• rotate the disk several times and combine the "duplicate" values (i.e. that measure the same point on the disk) e.g. by using the mean value. If I remember correctly I could expect up to $\sqrt n$ times better values given $n$ rotations (according to $standard\ error = \frac{standard\ deviation}{\sqrt{sample\ size}}$).
• rotate slower to get more precise measurements
• rotate slower to get more samples per turn (could also be factor $\sqrt n$ here)
• repeat the overal measurement several times and combine the results. Possibly another factor $\sqrt n$?

No options are:

• additional sensors
• different sensor
• modify the enviroment (control temperature, air circulation)
• linearization of the sensor

Are there any other options you know of?

• What kind of sensor do you use and which is the level of precision you are aiming for? Is the rotating disc flat? Does "breathing", "rolling" or "tilt" of the rotator matter and how about the heat (generated by the rotator and due to the environment)? All the factors $1/\sqrt{n}$ you mention above are the same: They are all due to the number of independent measurements. So you won't be able to reduce systematic errors. – Semoi Jul 20 '17 at 18:30
• @Semoi The sesnor is similar to this: tesatechnology.com/en-gb/products/… – Onur Jul 24 '17 at 8:19
• @Semoi The temperature will be compensated (assuming constant temperature), settings error and wobble errors can at least be partly removed. – Onur Jul 24 '17 at 8:25
• This is a contact sensor. It's linearity error is of the order of few microns so I expect that your main error contribution comes from the measurement against a rotating disc: The probe will jump and vibrate. Therefore, the individual data-points are not independent and the $1/\sqrt{n}$ factors not applicable. Rotating as slow as possible would in general be a good idea. I'm fascinated by your temperature compensation method :) – Semoi Jul 24 '17 at 19:44
• @Semoi constant temperature of the disc, i.e. no temperature gradient. The temperature itself may vary from part to part of course! – Onur Jul 25 '17 at 9:15

## 1 Answer

You can stop the rotation completely and take several (many) measurements to determine the inherent uncertainty in your detector. I don't know how much control you have on the sensor but this will help you determine if the sensor is accurate to a cm, mm, or better.

• This would take the "rotate slower" to the extreme... If I rotate in 1 minute or make stationary measurments shouldn't make a great difference I suppose. But I'll have a ty. – Onur Jul 24 '17 at 8:22