tip Calculating the Correlation Offset Separately for Each Capsule in a Condition

[John Cox]

The correlationOffset() Formula function can be a useful tool for identifying the time shift which maximizes the correlation between two signals: see this post for additional background. In some situations, a user will have a condition defined for time periods of interest (startups, process runs, specific product grades, specific modes of operation, etc.). The user then wants to analyze how the correlation offset varies for each time period of interest (each capsule in the condition). The key to this calculation is applying the transform() function to the condition in Seeq Formula, in combination with the correlationOffset() function. 

Let's say we have a temperature sensor in a reactor. At some point, well downstream of this temperature measurement, we have a relative humidity sensor that is sensing the same volume of air, but due to the locations of the two sensors, we know that the inverse correlation between the two signals is offset by a significant amount of time delay (at least 2 hours, as visually estimated with the dashed regions in the trend below).

 

As a reminder, for this use case, the objective is to calculate the correlation offset separately for the data contained within each capsule in a condition of interest (shown as Time Periods to Calculate Correlation in the trend below):

 

The formula approach for this is shown below, with comments to describe the details. The transform() function enables the correlationOffset() function to be applied separately to each capsule in the Time Periods to Calculate Correlation condition, and the correlation offset time is stored as a capsule property of the condition generated by the formula:

 

The resulting calculated offset (in units of seconds) is shown in the capsules pane at the lower right and also at the top of the screen as labels. Optionally, the "Offset" capsule property can be converted to a signal (see Max Correlation Offset in lane 2) for trending purposes, and here the units were converted to hours. Looking at the final results, the time shift which maximizes the correlation between the 2 signals varies between 2.1 and 2.5 hours over the 3 time periods of interest shown in chain view, and this variation may offer valuable insights to the user. The time shift is a negative value which means that the relative humidity (downstream signal) would need to be shifted to the left by that time amount to maximize its correlation with the temperature signal. 

 

This is the formula to create a signal for the max correlation offset, based on the "Offset" capsule property. In this example the time shift is more meaningful in units of hours, so we convert from seconds to hours:

 

Note that in this use case we wanted to calculate the correlation offset separately for each capsule in a condition. If the goal is to calculate the correlation offset over rolling window time periods, there are other functions in Formula expressly for this purpose, such as CrossCorrelations_timeShifts().