Compared to time-consuming spreadsheet methods, advanced analytics offers a more effective approach to the analysis and reduction of batch cycle times, says Nick Gigliotti, analytics engineer, Seeq...
Production rates in batch processes are determined by the total cycle time of the batch, so decreasing batch cycle time is critical. Typically, cycle time reduction projects start with a cycle time analysis where an engineer divides the batch into phases – Pump In, Temperature Heat Up, Reaction, etc.– and then calculates the duration of each phase for past batches. After this analysis is performed, process optimisation efforts focus on decreasing the duration of individual phases to decrease the total batch time.
These batch cycle time analyses are typically time-intensive and frustrating projects because they require engineers to copy historian data into spreadsheets, and then define the start and stop trigger for each phase. This process is tedious for several reasons. First, spreadsheets only show numbers and timestamps, so engineers can’t see the phases they’ve defined on a time series data trend. Second, engineers are limited by the spreadsheet’s mathematical functions, which quickly become unwieldy and unusable as the analysis becomes more complicated. Finally, while defining phases for one batch is difficult enough, accurately scaling that analysis out to all past batches is extremely time consuming.
Fortunately, there are new advanced analytics applications available for engineers to perform batch cycle analysis. Seeq, a self-service advanced analytics application, reduces the time requirements and difficulties of cycle time analysis, leading to faster insights to reduce batch cycle times and increase production.
When an engineer uses a spreadsheet to perform batch cycle time analysis, he or she starts with an arbitrary point in time and then moves forward to identify the triggers for the batch phases. For example, to find the catalyst addition phase, one might start on January 1 at 12am, and then use the spreadsheet and historian to find the next time the catalyst valve opened, indicating the start time of catalyst addition. The engineer then looks forward in time to find when the valve closes, indicating the end time of the phase. The engineer now has two timestamps in the spreadsheet for the start and end time of catalyst addition.
Next, the engineer opens their historian’s trending application, finds the two timestamps, and might find that the valve closed for only one second. The engineer knows this blip does not define the end of the phase, so he or she goes back to the spreadsheet and repeats the process of finding the end timestamp. The engineer can never see the phases defined on the spreadsheet without double checking the timestamps on the historian trend, a time-consuming exercise.
Seeq eliminates the guesswork in defining phases of a batch cycle with a feature called “capsules”. Users define a time period of interest, a capsule, for “Catalyst Addition” by using Value Search, a point-and-click tool, to find when the valve was open. The result is displayed as a capsule, indicated by the green bar at the top of the display pane in Figure 1 (top).
The engineer can also see the “blips” where the valve closes then reopens, each of which shouldn’t be included in the phase definition, and can then adjust a parameter in the Value Search to ignore short gaps. Using Seeq, engineers get immediate feedback as to whether this adjustment had the desired effect because the two capsules with a small gap between them are now joined as one (Figure 2, above), with the green line at the top of the figure representing the capsule now joined.
Defining batch phases is never as simple as finding when valves open and close. Engineers must also use process variables, like temperature or pressure, to define each phase. For example, in a reaction phase, there often is a Temperature Heat Up step where the temperature is raised to the reaction temperature setpoint. The engineer defines the Temperature Heat Up start time as when the temperature starts changing after pumping material into the reactor and the Temperature Heat Up end time as when the temperature stops changing before Catalyst Addition.
Using spreadsheets, the engineer must load an arbitrary amount of temperature data from the period after material has been pumped into the reactor and then calculate the difference in temperature at some interval. However, the engineer knows measurement noise can skew results, so he or she calculates a moving average first, then calculates the temperature differences.
Once the engineer has identified the times where the temperature starts and stops changing, he or she then verifies the timestamps are correct by referencing the trending application (as with the Catalyst Addition analysis). The engineer then finds that the temperature rises during Catalyst Addition as well, so the analysis must be repeated to not include Temperature Heat Up during Catalyst Addition. This approach takes a lot of time and repeated effort.
Seeq makes defining phases with process variables much easier as it can be done using point-and-click tools. First, an engineer can “smooth” the temperature signal using the Low Pass Filter tool, and then use Seeq Formula to calculate the derivative of the “smoothed” temperature signal. The point-and-click Value Search tool can then be used to find when the derivative is greater than 0.
Finally, a tool for combining capsules can be used to find when the temperature is changing when Catalyst Addition is not present. The Catalyst Addition capsule and Temperature Change capsules overlap, so temperature changes during Catalyst Addition should not be counted in the Temperature Heat Up phase definition.
In order to accurately perform cycle time analysis, phase definitions must be accurate for all past batches. Engineers working with spreadsheets define the phases for one batch, then replicate that analysis for other batches, often scaling out to hundreds or thousands of batches.
With Seeq, the analysis performed on a single batch is automatically performed on all past batches, simply by zooming out the trend display to include more time – from several hours to a week, month, or year. The engineer can sort the batches by duration to find the fastest batches, and use those for further analysis, such as golden profile development or statistical analysis.