Bosch Packaging Technology has introduced its new data mining service, which has recently been added to the range of pharma services for solid dosage forms. The service is offered by the Bosch subsidiary Hüttlin and aims to evaluate existing machine data more effectively to identify and eliminate root causes.
           
“So far, about 50 percent of deviations have been classified as ‘human error’,” said Dr. Marc Michaelis, expert for continuous production and process verification at Hüttlin. “Yet we assume that this is true for no more than ten percent. The rest is often misinterpreted due to missing information. As a matter of fact, there is already enough data available to get to the bottom of the causes. However, there is a lack of knowledge and time to read this data correctly.”
           
The data mining tool examines large amounts of data for the smallest effects using statistical methods. In general, the data from two production batches is already sufficient to draw first conclusions. The more data is available for evaluation over a longer period, the more details can be identified. All it takes is machine sensors, which already collect data on almost all historical machines, as well as the right tool to disclose the data.
           
“Large investments are not required to use the existing data more effectively. The key to success lies in merging knowledge from different disciplines,” said Dr. Michaelis. “Bosch not only has the necessary technical expertise, but also extensive process knowledge in customers’ product manufacturing. To identify reasons for process deviations, which are not obvious at first sight, we teamed up with the statistics experts from our Bosch Center for Artificial Intelligence in Germany and the U.S. Together we will raise the data treasure.”       
           
The potential of this approach has already been demonstrated successfully in different customer projects. For instance, when a customer suddenly produced a “out of specification” (OOS) batch, the Bosch experts systematically got to the bottom of things. The recorded data showed that a particular valve was responsible for the deviation. However, the valve had been excluded beforehand since it was considered uncritical in terms of product quality. Thanks to the data analysis, deeper correlations and an undetected cascade of connections could be identified. “Eventually, we found out that the valve provided an indirect indication of a false gas flow in the system, which was not visible at first sight. The problem could then be easily solved by recalibrating the system,” Michaelis explains.
           
By allowing for successful root cause analyses and process improvements the pharmaceutical industry can better maintain consistent high quality.