Umetrics Suite Blog

An example of how multivariate statistical process control improves batch quality in biopharma manufacturing

November 22, 2018

For pharmaceutical and biopharmaceutical manufacturers it’s common to generate and store large amounts of data originating from a variety of sources. This valuable asset has the potential to deliver critical benefits if the right applications are built around it. Data analytics tools are an important component in making that happen.

Umetrics bioreactor biopharma

MVDA provides a way to better understand data from a variety of batch processes and enables statistical process control.

Whether continuous or batch, parallel processes or serial steps, as data collection expands univariate analytics quickly become limited and inefficient. Multivariate techniques provide a more holistic and efficient way to model and create applications around the data — applications that may be used, for example, to detect variation and events in the processes.

Being able to monitor batch processes for quality and yield is another possible benefit of multivariate data analysis (MVDA). Also important is the ability to stabilize yield and other batch to batch quality parameters. Together these applications can have a significant impact on the production environment while supporting patient safety and a strong bottom line.

Further, Design of Experiments (DOE) and design space modeling in combination with MVDA for monitoring and prediction can support continuous process verification and allow a proactive and more flexible approach in validation of the production process. Continuous process verification has been described in ICH Q8 (R2).

From monitoring to prediction and control

MVDA can be applied in process monitoring and forecasting as well as for prescriptive analytics and control.

For process monitoring, a reference trajectory characterizing the evolution of the process is first established based on historical batches. These types of models are also useful for offline troubleshooting and increased process understanding. Straight-forward connectivity to process historians enables real-time multivariate process monitoring and early fault detection.

Predictive and prescriptive multivariate analytics, forecasting, and model predictive control provide means to predict the future of a process trajectory and suggest changes to parameter set points in order to optimize a certain process outcome and close the loop.

control advisor simca-online

Control Advisor in SIMCA-online provides real-time multivariate monitoring that includes forecast and advised future mode for process optimization.

Example from OSIsoft PI World

To better understand how a pharmaceutical company may put the historical data from their processes to good use to monitor deviations, apply real-time adjustments of the processes, and gain a positive impact on their ROI, we can consider the OSIsoft PI World Conference presentation from Janssen Pharmaceuticals.

Phuong Vo, a scientist from the Technical Operations department of Janssen Pharmaceuticals in the Netherlands, outlines an application of multivariate statistical process control (MSPC) using PI data and SIMCA-online in a biologics manufacturing environment. A specialist in Process Validation, Continued Process Verification and Life Cycle Management, Vo is also the project lead for the real-time multivariate data analysis (RT-MVA) of manufacturing upstream process at Janssen Leiden.

Like many manufacturers, Janssen Pharmaceuticals collects and stores large amounts of process data in multiple PI systems. Vo’s PI World presentation states: “It can be challenging to observe trends when there are 10+ variables impacting process performance. MSPC changed that. Multiple variables across batches are monitored in real-time to detect any developing trends that may affect the batch performance.”

Janssen’s approach to data management

The company initiated a project dubbed “Prometheus” to use the PI System and SIMCA-online to build a multivariate monitoring system for their production process. By gaining more insight into their process data using multivariate data analytics they were able to realize many benefits.

In her presentation, Vo stated that the company’s vision was to optimize the global deployment of the PI system with integrated data management for:

  • Continued process verification (CQA and CPPs)
  • Efficiency (batch reporting and automated data capture)
  • Real-time process visualization
  • Real-time process analytics

Janssen undertook the initiative to better serve a number of internal stakeholders as well as their customers. The project addresses the needs of these stakeholders including:

  • QA Representative – to be able to approve and release batches faster and more efficiently
  • Business Support – to be able to analyze all the process data for cycle time and performance
  • Maintenance Technician – to gain visibility on energy usage
  • Plant Manager – to gain visibility on all processes in real time centrally from his/her desk

The process involved the production of a monoclonal antibody. The production steps in the process include: Preculture > Fermentation > Direct Product Capture > Purification > Final Testing.

In the Fermentation stage, the data collected into the PI system comes from probes directly within the bioreactors. Altogether, there were more than 50 online parameters coming from the probes and more than 15 offline parameters coming from measurements taken, such as cell density, viability and titer.

With so many different parameters from disparate sources, the company was finding it a challenge to know how the process was performing. There was so much data, but not every engineer needed to look at all of the parameters, or the same ones. They needed a way to monitor a large amount of data in an efficient way and uncover process deviations more effectively.

current process monitoring Janssen PI World

Too many variables from multiple sources made it difficult to know what to look for.

The solution for Janssen was MVDA – being able to analyze multiple variables at the same time. Using MDVA, the company was able to get a better picture, creating a parameter to summarize all variables that can be monitored in real time.

mvda chart Janssen PI world

MVDA creates a summarized parameter with more relevant meaning.

Key Takeaways

Using PI data and SIMCA-online Janssen was able to create a simple, intuitive chart, providing an overview of all variables, taking into account the correlation between them. The team was able to drill down to relevant information in just a few clicks.

They were able to identify key MVDA signals that indicate process deviations, and understand what corrective actions could be taken to remain within the approved design space for regulatory compliance.

For Janssen, one of the key process deviation signals was online pH probe drifting.

Janssen Example Probe Drift Detection PIWorld


Another key process monitoring signal, was early detection of contamination.

Janssen Process deviation example PI World

Early detection of contamination saves thousands of dollars in the long-run by preventing contamination of downstream batches.

 Janssen early detection savings down stream batches

For example, the Janssen Prometheus trial showed results such as early detection of contamination that saved downstream batches, leading to a savings of up to €200,000.

After an initial successful trial period in Leiden, Janssen began global deployment of SIMCA-online for process monitoring.

Watch the presentation

“Use of PI and SIMCA-Online to Build a Real-Time MSPC System”
Janssen Pharmaceuticals (PI World) 2018



Download the presentation


Want to know more?

To learn more about analyzing batch process data for continuous process monitoring, register for this upcoming webinar.

Register for webinar


Topics: Pharmaceutical manufacturing, Statistical Process Control, Manufacturing Processes

Anna Persson

Written by Anna Persson

Anna Persson is the Senior Principal Data Scientist for Sartorius Stedim Biotech based in the Greater New York City Area. Anna is responsible for the online and off-line implementation of the Umetrics Suite of Solutions (SIMCA, SIMCA-online, Control Advisor, SIMCA-Q, MODDE, Easy Analytics, Active Dashboard) in customer environments. Anna is a graduate of Umeå University.

Leave a comment