2 logo-BSH

A team of Axiologo data scientists and consultants helped BSH to conduct a thorough data analysis that clearly exposed the main factors influencing the quality of the coffee grinders. Based on these a solution was developed – a coffee grinder digital twin.

1 BSH Naslovna





BSH factory in Nazarje is one of the most advanced serial manufacturing production sites in
Europe. It produces coffee machines for brands like Bosch, Siemens, Gaggenau, Neff and won the Factory of the year 2020 award.

One of the main components of the coffee machine is the grinder and its production process is complex. Many components and a vast range of production settings influence the
performance of the grinder and the taste of the end product – coffee.

At some point, the number of grinder quality issues increased and the engineering team was challenged to solve them quickly. They decided to try a new approach to address quality issues – a data-driven analysis and application of data science to solve them.





A team of Axiologo data scientists and consultants helped BSH to conduct a thorough data analysis that clearly exposed the main factors influencing the quality of the coffee grinders.
It also omitted some assumptions that the engineering team had.
Data revealed that the influence of different factors could be quantified and even the interactions between different factors could be estimated.


The result of the analysis phase was a clear understanding of what are the root causes of issues and how to address them to find a solution.


Using data science and modeling a coffee grinder digital twin was developed.
The digital twin is using Just-In-Time production data and processing it to define the optimal
setting of the production line to produce zero defects coffee grinders. Optimal production settings are calculated for each coffee grinder on the production line.







  • The production of coffee grinders in BSH Nazarje with the help of its digital twin is running
    with near ZERO DEFECTS attributable to the production process.

  • The amount of 100% testing regime (related to lower product quality) has been reduced by a factor of 6.



  • The workload of production technologists related to analyzing issues and finding solutions was
    significantly reduced and enabled them to focus on development instead of troubleshooting.

  • Most importantly the serial production of coffee machines is running smoothly and the production throughput of the facility has increased by 2%.



Pills-2 flip


Faster and more efficient drug development by optimal use of R&D data.





A global pharmaceutical leader was in the process of reinventing its drug development process to become faster and more efficient. One of the initiatives they were looking at was inSilico drug development – a data-driven approach to drug development that aims to maximize the use of data and data models to speed up the decision-making in the drug development process and shorten the whole process.

Their pharma research team was looking for a partner who could help them with advice on data and data science topics and at the same time support the business transformation process.


Using data science a multidisciplinary team of pharma scientists, data scientists, and business consultants was formed that led the initiative from its first step, the feasibility study, to a full scope implementation and roll out in multiple locations around the world.


One of the main deliverables of the project was the inSilico process – a new aspect of the drug development process that helps pharma scientists to:

  • Acquire and validate drug development data quickly

  • Analyze R&D data using best-practice analysis concepts and tools

  • Simulate drug behavior using data-driven models



The final product of this process is a pharma digital twin – a model that engrains most of the experience gained in the development process and can be used during the product life cycle to


  • Troubleshoot production issues

  • Simulate the impact of changes of ingredients or technological processes

  • Optimize production site transfers


  • Faster and more informed day-to-day decision making by pharmaceutical R&D teams

  • Standardized analytical tools allow team members to gain more insights

  • inVitro and inVivo models allow omission of certain physical experiments contributing to less effort and cost reduction

  • The inSilico process that links all the activities (data – analysis – modeling – inSilico experiments) enables the customer to speed up the drug development process and reduce costs by a significant factor.


Pills-2 flip


When a manufacturing process deviates from its expected range production, quality, environmental issues, and cost issues may arise.





A global pharmaceutical leader wanted to improve the management of the manufacturing process on LAB scale. Due to a high number of samples being manufactured, many formulations, and process variations a thorough analysis of the manufacturing process was done only after the process had already finished.

Sometimes critical process interruptions and anomalies have been identified at a point where they could not be fixed anymore and the manufacturing process needed to be repeated.

In a drug development environment where speed is key, this is a strong drawback.



A team of pharma and data scientists jointly analyzed a representative process data sample to define which types of anomalies occur and what process interruptions need to be detected.

Depending on the nature of anomalies different approaches to anomaly detection can be implemented. From simpler anomalies’ absolute changes, variability and standard deviation can be used to detect them, for more complex ones like multiparameter deviation, detection metrics can be used.

The validation of the solution has proved that more than 95% of process interruptions and anomalies could be properly detected. The solution processes all key drug manufacturing process parameters at the same time and provides researchers with a quick insight into which process aspects they need to review and validate.

In the future, the solution will be processing real-time manufacturing process data and will allow the researchers and operators to respond to process anomalies promptly – to address manufacturing process issues before the whole batch is wasted.



  • Better insight into process data: Algorithms do the job of an expert to examine process data and point to anomalies and other process deviations. Data is examined to a greater extent.

  • Speed: Processing can be done in real-time and operators can automatically be notified when an anomaly occurs. With the use of predictive models, anomalies can even be predicted and prevented.

  • Efficiency: The cost of processing all manufacturing data is a fraction of what it would be if a manufacturing process expert would do it.