Averaged vs Raw Data Analysis

1. Averaged vs Raw Data Analysis

This article shows the shortcomings of using averaged data instead of raw data for operational analysis. Averaged data is typically fine for things like production variables — production reports, material and energy balances, and so on. However, averaged data should not be used for monitoring the performance of a process or equipment, condition-based maintenance, event monitoring, or environmental exceedance.

The analysis below was conducted using three datasets from a customer's process:

• Tank level
• Tank pressure
• Chill water temperature

The trends below show the raw data and 1-minute average data for a period of 4 days and 14 hours. The 1-minute average data was further split into three datasets:

• 1-minute time-weighted average stored at the end of the averaging period
• 1-minute time-weighted average stored at the start of the averaging period
• 1-minute event-weighted average stored at the start of the averaging period

All three 1-minute average datasets give similar results, so the rest of the analysis focuses on the 1-minute time-weighted average stored at the start of the averaging period.

Zooming into the chiller temperature trend to a 15-minute window, you can see the four different datasets. The averages generally follow the raw data reasonably well when the raw data is steady, but start to vary significantly when the process varies.

Below are trends for the three process variables over time ranges where the averaged data significantly deviates from the raw data.

Below is an example for tank pressure, where the % error is in excess of 30%.

The next step is to calculate the % error (averaged value vs. raw value) for each process variable, then assess how many times this % error is larger than 1% during the 4-day, 14-hour period. Below is a table of the % errors for the three process variables, sorted largest to lowest.

The total number of % errors of 1% or more over the period is shown below. Note the highest error for temperature was 3.2% because temperature varied more slowly than the other two parameters. For pressure, the highest error was over 115%.

Conclusion

This analysis clearly illustrates the inadequacy of using averaged data to monitor process operations. Averaged data can carry large error and can easily miss significant process deviations that should be flagged — deviations where Event Frames should be created to mark them and take action.