Algorithm Protects Active Pharmaceutical Ingredients

Download (377 KB)

Executive Summary

Vetter Pharma-Fertigung GmbH & Co. KG

Customer profile

Vetter Pharma’s core business is the manufacture and packaging of aseptically prefilled syringes, cartridges and vials. Around 5,700 people are currently employed at around a dozen production and sales sites in Europe, America and Asia. Key tasks at the Ravensburg Vetter West site include manual and automated visual quality control of the products and logistics. Customers – pharmaceutical and biotech companies from all over the world – as well as national and international regulatory authorities place the highest demands on compliance with GMP requirements. For more information, please visit

Outline of the project

Vetter Pharma-Fertigung GmbH & Co. KG. collaborated closely with Yokogawa to develop an early detection of freezer defects. By monitoring over 300 freezers, critical temperature trends can be reliably identified, giving staff sufficient time to react to conspicuous events. The detection system fits seamlessly into the existing system landscape and is configured in such a way that it can be transferred to other locations without great effort. The adaptive algorithm works with a single set of parameters for freezers of different types and temperature ranges.


The Challenges and the Solutions

The Challenges: Safe operation and early maintenance

Freezers play a crucial role at Vetter Pharma in Ravensburg. Worthy Active Pharmaceutical Ingredients (APIs), primarily biologicals, are stored there - at temperatures as low as -80°C - waiting to be processed or shipped all over the world. The company operates more than 300 freezers at the Ravensburg Vetter West site alone. Malfunctions of the freezer function must be detected as early as possible before the failure so that the valuable contents can be relocated in time if necessary. Of course, temperature sensors in conjunction with dedicated control systems for room monitoring have always continuously watched over the freezers as well. However, this was only done by means of high and low alarms, so that after a high alarm there was only little time before the frozen APIs left the permissible, often very narrow temperature range. Due to extremely high quality and safety requirements, it would then have to be disposed in the worst case.

The Solution: AI-driven anomaly detection

Observations by the experts at Vetter Pharma also already led to an important lead. Michael Kratzmann, Head of Production Engineering at Vetter Pharma, explains: "Spontaneous failures of the freezers 'out of the blue' are extremely rare. Usually, a malfunction announces itself in the form of anomalies in the temperature curve, but without exceeding the alarm limits." Under normal conditions, the temperature fluctuates periodically around the set point with constant amplitude. Significant indications of an imminent, critical disturbance are given by a sudden transition to a non-oscillating, linear temperature course as well as preceding changes in amplitude and wavelength of the oscillation.

Based on Vetter Pharma's findings, AI specialists from Yokogawa began systematically analyzing around 700 GB of measurement data from the PIMS as part of a feasibility study in spring 2020. In the process, numerous further deviations from the normal state were found. The intensive dialogue with Vetter Pharma yielded important clues as to how these should be evaluated. Most of the deviation patterns, according to the result, do not, or at least not immediately, point to an imminent malfunction. Dr Silke Müller, Solution Consultant Data Analytics at Yokogawa Deutschland GmbH, summarizes: "The essential requirement for our algorithm is to detect a significant, critical temperature pattern and to evaluate other deviations as non-critical in order to achieve a high detection rate and thus reliability with an overall low alarm frequency."
The actual implementation project began in the summer of 2020. The room monitoring control system acts as the real-time data source, from which temperature data of all connected freezers are continuously transmitted from an operating station via an OPC interface data unidirectionally. Additional data sources, such as paperless recorders used at other Vetter Pharma sites, can also be integrated via OPC.
All data is stored in the CI server, processed by the algorithm, the results interpreted, processed and visualized. The alarms are then sent via OPC DA using Vetter Pharma's existing central alarm system, a Siemens WinCC installation. From there, the alarms reach the responsible production managers or technicians via different, systematically defined communication channels. All this takes place in a decentralized architecture, so that control rooms for room monitoring are superfluous.

The Challenges and the Solutions


The Results

Although freezers of different types in different temperature ranges are in use at the West site, a single algorithm with a single set of parameters is sufficient to monitor them reliably. A total of 310 freezers have been connected since the system was finally commissioned in December 2020. Only seven freezers of special design are currently not yet integrated. For them, the temperature oscillation amplitudes are well below one kelvin and are thus too small to be measured precisely and detected reliably by the algorithm.
Basically, the periodic fluctuations of the temperature data are converted via Fourier transformation into a signal that can then be analyzed quantitatively. In a training phase lasting several hours, the algorithm first observes the temperature curve of each freezer that has been newly commissioned or whose status or operating point has been changed. Even during the subsequent diagnostic phase, the algorithm automatically refines its detection potential.
The result impresses Matej Dragonja, the project manager at Vetter Pharma: "In a period of two weeks, the system typically delivers four alarms from all connected freezers, two of which remain without recognizable consequences. Two alarms are critical and require either further checks or the transfer of the freezer contents to a reserve freezer. We keep dedicated staff on hand for this at all times." While a warning time of one hour is usually sufficient for safe relocation, the new diagnostic system offers a safety buffer of six to eight hours according to previous experience. This relieves the staff even if several critical events should occur almost simultaneously.

The Results


Customer Satisfaction

With the AI-supported monitoring system, Vetter Pharma has entered new territory. "So, it goes without saying that the fine-tuning of the event evaluation had to be done step by step, as we didn't want to take any risks. All of us, both in engineering and in production, are very satisfied with the solution. It fits seamlessly into the existing system landscape, works reliably and gives us a high degree of additional safety with minimal operating effort," Kratzmann summarizes his experience.

Silke Müller emphasizes another aspect of the project: "From the kickoff to the feasibility study, the implementation and initial parameterization of the algorithm to the commissioning, we were not on site in Ravensburg. The project was - not least due to corona - completely coordinated and carried out remotely.“

Related Products & Solutions

  • Digital Transformation

    With Yokogawa's wide knowledge of and expertise in process manufacturing, Yokogawa can help bring about the realization of a digital transformation that will lead to a better future for its customers.

    See More