Advanced Tech Elevates Process Modeling in Batch Processing Industries
Advanced Tech Elevates Process Modeling in Batch Processing Industries
Advanced Tech Elevates Process Modeling in Batch Processing Industries
© Just_Super/Getty Images
June 2022
Pharmaceutical, consumer health, and consumer In addition, the quality and R&D functions can
packaged goods companies have worked for upgrade product and process parameters during
decades to achieve an expert-level understanding the next R&D cycle.
of their end-to-end manufacturing processes
using various modeling techniques. Traditionally, This article describes advanced process modeling,
process modeling has relied almost exclusively introduces three use cases for it, and offers
on human experience and expertise. Now, with the guidelines to boost the likelihood of success
help of newer technologies such as machine with its use.
learning, predictive modeling, and faster data
processing, process manufacturers can
build more advanced models that go further to Upgrading business process modeling
optimize processes even before producing Process modeling is a manufacturing discipline
a batch, forestalling problems likely to occur for understanding the effects of process inputs (raw
during batch production. materials) and CPPs on final products. The more
adept manufacturers are at process modeling, the
Companies can leverage these models to improve better able they are to monitor and control the
performance and quality—even beyond what CPPs that determine a product’s critical quality
regulators may require—and save time and resources. attributes (CQAs). CQAs are the physical, chemical,
Leading companies that use this tech-enabled biological, or microbiological properties or
approach to process modeling have seen a reduction characteristics that should be within an appropriate
in deviations (for example, products outside limit, range, or distribution to ensure the desired
specifications) of more than 30 percent and have product quality.1 In the case of shampoo, for example,
cut the overall cost of quality by 10 to 15 percent. pH balance is a CQA. In pharma, nearly all CQAs—
including, to name just two, concentrations of active
In addition to these tangible benefits, advanced principles and excipients—are highly regulated.
process modeling deepens manufacturers’
understanding of their production processes. With Process modeling informs the controls strategy that
these insights, plant operators can adjust critical defines what, when, and where to test to ensure
process parameters (CPPs) such as temperature or that the process performs as it should and maintains
pH and improve decision making during production. the quality of the product. Controls may include,
1
Guidance for industry: Q8(R2) pharmaceutical development, revision 2, US Food and Drug Administration, November 2009.
Exhibit 1
Product mastery
Product mastery models
models drive
drive impact
impact through
through specific
specific use
usecases.
cases.
Capabilities required for use of advanced process models, level
High
Medium
Low
~70% of impact ~30% of impact
One-off changes Set-point optimization Continuous use
Model identifies significant process Model analyzes variability of Online model continuously
improvements and tests potential inputs from batch to batch for receives process inputs,
hardware and process adjustments recipe adjustment predicts outputs, and may
prescribe corrective actions
Analytics
Use becomes less prevalent as Ongoing use involves heavy Typically, already optimized
process matures data science and problem- levers are optimized more
solving, eg, on other products frequently
Information
technology/ None Can typically be done in Requires automated data flow
operational current setup and model run supported by
technology information- or operational-
technology workstream
Selected Adjustment of reactor filling level; Recurrent but infrequent Real-time optimal parameter
examples changing valve settings reruns of the models based on suggestion
offline data, eg, simple tank
Addition of buffer capacity farm planning Full closed-loop approach
Exhibit 2
Machine learningmodels
Machine learning models identify
identify the
the biggest
biggest variability driversininquality
variability drivers qualitytest
test
results,
results, and
and simulate
simulate effects
effectsofofadjustment.
adjustment.
An automated, real-time data pipeline feeds a quality-control (QC) model
QC dashboard, illustrative
Drivers of variation
Compression machine speed Coating room air pressure Assay QC test result
Exhibit 3
Soft
Soft sensors
sensorsenable
enablein-process
in-processoptimization,
optimization,increasing
increasingthroughput
throughputby
byup
up to
to 15 percent.
15 percent.
Adjustment of process conditions based on product predictions
–100 0
July Oct Jan Apr July 100 120 140
The US FDA classifies process models based on impact, and applies different
The US FDA for
requirements classifies process models based on impact, and applies
each type.
different requirements for each type.
FDA requirements, by model type
Description Low-impact models Medium-impact models High-impact models
point will necessitate changes to different quality understanding to the next level. But it also requires
strategies such as the validation strategy and the that companies build in-house expertise, finding the
in-process control strategies. It is therefore right balance between off-the-shelf commercial
crucial that the quality organization play a major role software and tailored solutions. Understanding the
in developing process models, so it can adapt technology is crucial for building trust in the model’s
accordingly and properly explain the new models recommendations, which in turn will increase the
and related tools to regulators (Exhibit 4). users’ understanding of the manufacturing process.
Only then can manufacturers unlock the full potential
of advanced process modeling.
Álvaro Carpintero is a partner in McKinsey’s Madrid office, where Miguel Ángel Morán is a senior expert; Evgeniya Makarova
is a partner in the Chicago office; Matthias Spiegl is a partner in the Hamburg office; and Vanya Telpis is a director of
knowledge in the New Jersey office.
The authors wish to thank Helena Lubiano for her contributions to this article.