We released our Batch Analytics (BA) application with version 12 of DeltaV and provided some additional enhancements as part of version 13. BA uses Multivariate Analysis and Dynamic Time Warping to detect process faults, the reasons for those faults, and predicts endpoint quality, all in real time. So instead of having to wait until the batch completes to find out there was a problem, fault and quality issues can be examined while the batch is still running. This allows operations and engineering personnel to make better decisions that could correct a quality issue, dump a bad batch early, or schedule maintenance for when a unit is not in use.
Another important benefit is the education of inexperienced personnel to gain process understanding. One of the features of the fault detection screen within BA is to prioritize parameters that are contributing to a fault:
The small green band at the bottom of the screen is the normalized range of the two fault parameters, T2 and Q. A fault in this range (0 to 1) is statistically insignificant. The larger the fault peak (the large blue T2 peak is around 55), the more statistically significant the fault is. By selecting the user-friendly parameter names on the left, response plots of actual versus modeled are displayed:
The black lines above are the actual parameter response, while the dashed and dotted blue lines are the expected, modeled response. For instance, you can see that M1 Level didn’t increase as much as the model thought it should have and the Salt Bin Level didn’t drop as much as the model predicted it should have.
But you don’t have to have a fault to monitor actual versus modeled trajectories. Here’s a fault detection screen from a normal batch:
You’ll notice that while some of the peaks exceed the normalized 0 to 1 range, the largest fault is less than 2.5, compared to the 55 on the previous example. My point is you can still see the parameters on the left hand side and can plot out their actual versus modeled response:
Notice that the mixer and salt bin levels followed the predicted, modeled response. This can provide a great tool for training operations personnel to understand what “normal” is to be better prepared for process upsets and faults.