HOW CAN I MAKE SENSE OF ALL THIS DATA?

摘要

With the advent of DCS, systems mills have the opportunity to gather "mountains" of data. The data helps define the process; however, sometimes there is so much data that it doesn't make sense and we are unable to "scale" the mountain. The following are some techniques I have learned to "scale the mountain": 1. Is there a trend in the data? Using rolling averages from 3 to 30 data points generally helps to determine if a pattern exists. Rolling averages can help pinpoint problematic periods and potential relationships between different parameters. 2. How can I predict BODs? The relationship between a short-term test and BOD5 generally exists but can change, and it is best to update the relationship on a rolling 7 to 14-day basis. The relationship is updated whenever a BOD5 test comes out. 3. How much does a change in one parameter impact another? Group the data based on ranges of the dependent variable. The average of independent and dependent variables for each range of the dependent variable is graphed to determine the slope or impact of the independent variable on the dependent variable. 4. How stable is my operation? Developing a probability distribution for the data and graphing it on probability paper will help define the stability. The steeper the slope within the 5% to 95% probability generally depicts stability for 90% of the time, while the length of the tails from 95% to 99.9% (or 0.1 to 5%) indicates the limits of the instability. 5. What if the process goes awry? — Compare process data from when the process was performing well versus when it was performing poorly. Good vs. bad comparison helps to point out the problematic areas of the process.