Recently, my colleagues and I worked with a BioPharmaceutical  Manufacturing company to apply predictive analytics methods to solve a  critical product quality problem. We were looking for root cause(s) of this  complex problem.

 The company manufactures vaccines, under strict regulatory guidelines for  sterile manufacturing and packaging environments.

 The Problem: Specifically, in an important part of the process in which  they manufacture one of the active ingredients for a vaccine, they were  scrapping about 30% of batches.

 Conservatively, scrap at this level resulted in millions of dollars in  opportunity cost (i.e., there was unmet demand for the vaccines), lost time  and resources since it takes several weeks for each batch to be  manufactured, and materials expenditures. The problem was complicated  by the fact that the manufacturing process was complex with:

• a sequence of several interdependent steps
• many raw materials added to the process from different vendors and lots
• the process showed trends across subsequent batches

Here is a data visualization of the problem.

How They Tried to Solve the Problem: This company had a very competent engineering staff, with numerous PhD BioChemical Engineers. They had an in-depth knowledge of their process from raw materials through to finished product testing and release.

These same engineers also had years of anecdotal evidence about how the process behaves. They had conducted numerous brainstorming sessions to resolve the quality issue with engineering methods, looking at what raw materials and process variables could contribute to out-of-specific outcomes based on the underlying science of the process. They had tried manipulating small numbers of raw materials and process variables in designed experiments, based on this engineering approach of what the underlying science would indicate.

These approaches did lead to some insights and improvements but none moved the needle far enough to resolve the problems and its associated costs and visibility within the organization.

Root Cause Analysis: The next blog will cover how predictive analytics (data mining) was used for root cause analysis. It will discuss the hidden patterns. The scrap rate is predicted to be about 5% after some process improvements are implemented.

Image Credit: http://www.flickr.com/photos/baronbrian/16725300/

Read other blog posts from Statistica:

	View All Recent Entries
	Data Visualization
	Enterprise Analytics
	Hidden Patterns
	STATISTICA
	STATISTICA Blogs
	STATISTICA News
	STATISTICA Visual Basic