Every control engineer knows measurement noise is a problem, but not all noise is the same, and the wrong fix can make things worse. This episode digs into how noise distorts what you think your process is doing, and what to do about it.

We cover:

• The filtering tradeoff. First-order filters, statistical filters and kernel filters each handle noise differently. Every one of them trades noise reduction against lag or delay.
• Noise that isn't random. Asymmetric noise can bias your averages. Autocorrelation makes a persistent disturbance look like a real trend. Low-resolution signals can look like noise, or hide it.
• Outliers aren't noise. Why spurious signals need a different tool than noise does.
• Noise or a real process change? When to filter, when to fix the source, when to adjust sampling rates, and why you shouldn't try to control noise your controller can't correct.
• Calibration and data reconciliation. Filtering doesn't fix measurement bias, so keeping instruments honest matters.
• What operators see. Why showing them the raw signal, not the filtered one, can be worth considering.

Follow along with the graphics and charts. This conversation is easier to follow with the figures in front of you. Open the three articles in R. Russell Rhinehart's "Develop Your Potential" series in Control Global's Measure section:

1. How to prevent noise from degrading control performance: filtering methods and their tradeoffs
2. Understanding asymmetric noise and autocorrelation in process measurements: noise that breaks the textbook assumptions
3. How can controls engineers distinguish noise from real process changes?: telling real changes from noise