WBF is holding their annual North American conference this week in Austin, Texas. For those not here, you can get a flavor for what’s happening through the WBF Twitter account and the search hash tag, #WBFna.
Two Emerson presenters are on the agenda today. First, Dawn Marruchella is teaming up with Lubrizol’s Robert Wojewodka to present, Benefits Achieved Using Online Analytics in a Batch Manufacturing Facility. Their abstract:
Batch operations present manufacturers with a unique setting where operators must work in a highly complex, highly correlated and dynamic environment each day. They must also manage a large amount of data and information on a running unit – all of this making it easy for batches to end up with undesirable processing events and/or less than desirable end of batch quality. Lubrizol wanted to improve their operations by providing their operators with the ability to detect upset conditions before they have a negative impact on their batches. In order to do so, they are collaborating to develop and deploy the use of online data analytics, based on multivariate analysis, initially at their facility in Rouen, France.
Here are some of my live-blog notes from their presentation. Operators and engineers work in a highly complex, highly correlated and dynamic environment and need to manage a large amount of data and information on a running unit. They need to avoid undesirable operating conditions and reduce variation, improve throughput and improve quality yet maintain safety. Data is everywhere and needs to be understood to achieve these plant objectives.
Lubrizol and Emerson jointly worked to develop viable on-line multivariate batch process data analytics to predict product quality on-line and on-line process fault detection and identification. Through a field trial, they wanted to document the benefits of this approach and learn about improvement opportunities.
Batch processes have challenges around process holdups, variations in feedstocks, access to lab data, and varying operating conditions. There is variability between batches, which makes analysis difficult. Borrowing from voice recognition technology, the team used Dynamic Time Warping (DTW) to characterize and align batch-to-batch comparisons.
By looking at the multivariate relationships, these questions can be asked about the running batch:
- Is it in multivariate statistical control?
- Is it within acceptable variation?
- Are any relationships atypical?
- Is end-of-batch quality still predicted within specification?
- Is there something I should be looking at regarding the health of the batch?
- Is there a way to get at what I need to look at very quickly?
The statistical methods used include Principal Components Analysis (PCA), Projections to Latent Structures (PLS), and PLS with Discriminant Analysis (PLS-DA). Dawn shared how the operators can see multivariable relationships trending in real time and flagging anything trending outside the norm.
The goal is not to have close loop control, but rather to provide process relationships for the engineers and operators to better understand how their process really operates. They do not want it to be a black box with the answers. Instead, it provides multivariable data relationships that cannot be seen. Here’s an article, Data Analytics in Batch Operations that describes this project and its results in more detail.