Models can help bring understanding to complex processes. In batch manufacturing, one of the benefits from generating models from process history data is the process understanding gained by going through the steps of generating and refining the model. These steps involve reviewing the data to be used in the model as well as examining the results of the model generation.
I caught up with Emerson’s David Rehbein, a Senior Industry Consultant with the Life Sciences industry team. You may recall Dave from blog posts about consultations with pharmaceutical and biotech manufacturers on applying batch analytics and optimization strategies. He shared examples from field trial and beta test sites on the use of batch analytics to aid in process understanding.
Using the Batch Analytics (pdf) software, once the batch process data is imported into the modeling tool, the data for all or selected batches can be viewed for a specific parameter.
This figure illustrates how the “B1 Level Indicator” is viewed for four batches.
In one of the test sites, the project team determined that the operators were performing a manual operation differently. This became evident by looking at the process parameter across multiple batches. All the operators started at the same process value and ended up at the same value. However, some used three large step changes to get to the final value while others used as many as eight. While this did not have an adverse effect on product quality (one reason this went unnoticed), it did have an economic impact on utility usage.
There was another example where an issue with a trigger that initiated a process step was discovered. Again, this did not impact product quality but did cause an overall reduction in line efficiency.
In another case, it was determined that the pH meters on the feed water to the process were out of calibration, which resulted in control problems.
Once a model is generated, process knowledge can be extracted even before the model is deployed. This can be achieved in several ways. One way is to add a batch that had poor quality to the test set for the model. By regenerating the model with this batch, you are able to view the faults that may have gone undetected that would have impacted the batch quality.
In this fault detection example, there was a fault detected with the greatest contribution to the fault from the “M1 Level Indicator”. By selecting that parameter from the list, a new window is displayed showing the history of that parameter during that stage. One can see that the level (dark line) did not reach the value expected by the model (blue line) for that stage of the process.
Applying these models as an offline analytical tool can help uncover process, instrumentation, and control strategy issues even before deploying the model in real-time. This not only enables improvements to the efficiency of the production process, it provides the user a deeper understanding of the factors that impact final product quality.
You can connect with the Life Sciences team through their Ask an Expert page.